Riksdagsmonitor Intelligence Platform โ€” API Documentation - v1.0.102
    Preparing search index...

    Hack23 Logo

    ๐Ÿ”ฎ Hack23 AB โ€” Riksdagsmonitor Future Threat Model

    ๐Ÿ›ก๏ธ Proactive Security for the Three-Horizon Architecture Evolution (2026โ€“2037)
    ๐Ÿ” STRIDE โ€ข MITRE ATT&CK โ€ข AI Workflow Expansion โ€ข Advanced Dashboards โ€ข Real-Time Data โ€ข AWS Serverless โ€ข Bedrock โ€ข Neptune โ€ข Aurora โ€ข Cognito

    Owner Version Effective Date Review Cycle Threat Scenarios Security Controls

    Horizon Horizon 1 Horizon 2 Horizon 3 Public Data Only Democratic Integrity GDPR Art. 9

    ๐Ÿ“‹ Document Owner: CEO | ๐Ÿ“„ Version: 2.1 | ๐Ÿ“… Last Updated: 2026-06-02 (UTC)
    ๐Ÿ”„ Review Cycle: Quarterly | โฐ Next Review: 2026-09-02
    ๐Ÿข Owner: Hack23 AB (Org.nr 5595347807) | ๐Ÿท๏ธ Classification: Public


    Establish a forward-looking threat model for Riksdagsmonitor's three-horizon architecture evolution (2026โ€“2037), covering new capabilities and expanded attack surfaces across the planned roadmap. This document complements the current THREAT_MODEL.md by analyzing threats specific to planned features that do not yet exist in production, and is the security counterpart to the strategy expressed in FUTURE_ARCHITECTURE.md, FUTURE_DATA_MODEL.md, FUTURE_FLOWCHART.md, FUTURE_STATEDIAGRAM.md, FUTURE_SWOT.md, FUTURE_MINDMAP.md, FUTURE_SECURITY_ARCHITECTURE.md and FUTURE_WORKFLOWS.md.

    ๐Ÿ“ Coverage dimensions (v2.1):

    Category Scenarios Controls Diagrams
    ๐Ÿ”ง Technical Security (STRIDE, ATT&CK) F1โ€“F12 FUT-001โ€“FUT-022 4 mermaid diagrams
    ๐Ÿ—ณ๏ธ Democratic Integrity & Accountability F13โ€“F16 FUT-023โ€“FUT-027 2 mermaid diagrams
    ๐Ÿ”’ Privacy & GDPR (H3) F17โ€“F18 FUT-028โ€“FUT-029 1 mermaid diagram
    ๐Ÿ”— Supply Chain & AI Governance F19โ€“F21 FUT-030โ€“FUT-032 1 mermaid diagram
    ๐ŸŒ Geopolitical & FIMI Cross-cutting Source-grading, FIMI detection 1 mermaid diagram
    ๐Ÿค– AI/LLM (OWASP Top 10) Cross-cutting Model-level controls โ€”
    ๐Ÿ•ต๏ธ Political-Intelligence Capabilities PI-T1โ€“PI-T7 Integrity-by-construction โ€”
    Horizon Versions Window Architecture Posture Dominant Threat Themes
    ๐ŸŸข H1 โ€” Static Baseline v1.x Today Static HTML/CSS on GitHub Pages, build-time agentic newsroom Supply-chain & CI/CD compromise, prompt injection, content integrity (covered by THREAT_MODEL.md)
    ๐Ÿ”ต H2 โ€” Static, Go Deeper v2.0 2026โ€“2027 Still static, richer pre-compute, CIA pipeline, party/OSINT analytics, real-time read-only feeds, 14-language translation Pipeline cache poisoning, multi-workflow AI orchestration, real-time data manipulation, translation integrity
    ๐ŸŸ  H3 โ€” AWS Serverless AI v3.0+ 2027โ€“2037 Amazon Bedrock, Neptune Serverless, Aurora Serverless v2, OpenSearch/Timestream/DynamoDB, AppSync/API Gateway, Cognito, Lambda, Step Functions, SageMaker, multi-region Cloud IAM & identity attacks, RAG/Knowledge-Base poisoning, graph/relational data exfiltration, agentic excessive agency, model manipulation, multi-region failover abuse

    Horizon boundaries are roadmap intent, not commitments โ€” the platform deliberately stays static through 2027 (FUTURE_ARCHITECTURE.md ยง2.6). H3 threats are pre-modeled so that controls are designed before the first managed service is provisioned.

    Aligned with Hack23 AB Threat Modeling Policy and Secure Development Policy.

    Planned Feature Horizon Target Date Architecture Impact New Attack Surface
    CIA Data Pipeline Integration H2 Q2 2026 Automated nightly fetch of 19 CIA visualization products External API dependency, data validation, cache poisoning
    Advanced AI Content Pipelines H2 Q2-Q3 2026 Additional agentic workflows (committee reports, motion analysis, week-ahead) Expanded prompt injection surface, multi-workflow orchestration risks
    Real-Time Voting Dashboard H3 2028+ WebSocket/SSE for live parliamentary voting data (requires Kinesis streaming backend) Real-time data manipulation, WebSocket security, connection state attacks
    Politician Profile Pages H2 Q3 2026 Per-politician detail pages with historical data Data accuracy attacks, profile defacement, SEO poisoning
    Enhanced Chart.js/D3.js Dashboards H2 Q2-Q3 2026 5 placeholder dashboards activated (Budget, Voting Patterns, Committee, Regional, Historical) Dashboard data injection, chart rendering exploits, large dataset DoS
    Automated Content Translation H2 Q3 2026 Machine translation pipeline for 14 languages Translation manipulation, cultural sensitivity attacks, LLM hallucination in non-English
    EU Parliament Cross-Reference H2 Q4 2026 Integration with European Parliament MCP Server Cross-platform data integrity, new external API dependency
    Bedrock AI Content Engine H3 2028โ€“2029 Step-Functions-orchestrated Lambda + Bedrock (Claude Opus, Nova Premier, Polly) article/image/audio generation Managed-LLM prompt injection, insecure output handling, excessive agency, model supply chain
    SageMaker Predictive Analytics H3 2028 Q4โ€“2029 Q1 Election forecasting, coalition & MP-vote models (SageMaker Serverless Inference + Feature Store) Training-data/feature-store poisoning, forecast manipulation, inference DoS
    Neptune Knowledge Graph + Bedrock Knowledge Bases H3 2029 Q2โ€“Q4 Semantic intelligence: graph (openCypher/Gremlin) + RAG vector search over 109K+ documents Graph-query injection, RAG/Knowledge-Base poisoning, embedding inversion, natural-language-query abuse
    Aurora Serverless v2 / OpenSearch / Timestream / DynamoDB H3 2027โ€“2028 Managed relational, full-text, time-series and NoSQL stores behind Lambda resolvers SQL/NoSQL injection, broken object-level authorization, data exfiltration, encryption-key misuse
    AppSync GraphQL + API Gateway Public API H3 2027โ€“2028 Managed GraphQL (real-time subscriptions) + REST usage plans GraphQL abuse (depth/complexity), broken authZ, subscription hijacking, API-key/usage-plan abuse
    Amazon Cognito Identity H3 2027โ€“2028 Authenticated user accounts (saved searches, alerts, personalization) โ€” first non-anonymous tier Account takeover, token theft/replay, IDOR, MFA bypass, GDPR scope creep
    AWS Amplify Web PWA + Native Mobile Apps H3 2028+ iOS/Android apps + PWA via Amplify, CloudFront + WAF + Shield edge Mobile API-key abuse, insecure local storage, certificate-pinning bypass, push-notification spoofing
    Conversational AI (Bedrock Agents, Lex, Transcribe/Polly) H3 2028+ Chatbot, voice interface, personal briefings, multi-agent autonomous tasks Agentic excessive agency, indirect prompt injection via voice, tool-chaining abuse, hallucinated political guidance
    Multi-Region Resilience (Aurora Global, DynamoDB Global, S3 CRR, Route 53) H3 2028+ Active-passive multi-region failover, global tables, cross-region replication Failover/route hijack, replication tampering, split-brain integrity, regional IAM drift
    Nordic & EU Federation (DK/NO/FI + EU Parliament) H3 2027โ€“2030 Shared data-mesh comparative analysis across parliaments Cross-jurisdiction data-integrity, source-spoofing, federation trust-boundary attacks
    timeline
    title Riksdagsmonitor Threat Landscape Evolution (2026โ€“2037)
    section ๐ŸŸข H1 โ€” Static Baseline (Today)
    2024-2026 : Supply-chain & CI/CD compromise
    : Prompt injection in agentic workflows
    : Content integrity attacks
    : GitHub Actions credential theft
    section ๐Ÿ”ต H2 โ€” Static Deepening (2026โ€“2027)
    2026 Q2 : Pipeline cache poisoning
    : Multi-workflow AI orchestration abuse
    : ๐Ÿ—ณ๏ธ Neutrality erosion via AI drift (F13)
    : ๐Ÿ”— MCP ecosystem compromise (F21)
    2026 Q3 : Translation manipulation (14 languages)
    : ๐Ÿ“Š Election forecast weaponization (F14)
    : Real-time data manipulation
    2026 Q4 : Nordic federation trust attacks
    : Cross-platform data integrity
    : ๐ŸŒ FIMI information laundering (F16)
    section ๐ŸŸ  H3 โ€” AWS Serverless AI (2027โ€“2037)
    2027 : ๐Ÿ”’ GDPR Art. 9 scope expansion (F17)
    : Cloud IAM & identity attacks
    : SQL/NoSQL injection
    2028 : RAG/Knowledge-Base poisoning
    : ๐Ÿง  Foundation model regression (F19)
    : Agentic excessive agency
    : ๐Ÿ“ฑ Mobile API abuse
    2029-2037 : โš–๏ธ EU AI Act compliance (F20)
    : Multi-region failover hijack
    : Post-quantum cryptographic transition
    : AGI-era governance challenges

    Dimension Current Future Rationale for Change
    ๐Ÿ” Confidentiality Public Public + limited Internal (H3) Platform content stays public; H3 Cognito introduces authenticated user accounts whose saved searches/alerts reveal political interest (GDPR Art. 9) and must be protected
    ๐Ÿ”’ Integrity High Critical Real-time voting data, expanded AI content, and H3 authoritative managed stores (Aurora/Neptune/forecasts) increase integrity requirements
    โšก Availability High Critical Real-time dashboards and H3 public API / multi-region services require higher availability during parliamentary sessions and election windows

    Note: This table describes the future Riksdagsmonitor system security classification. The CIA classification badges in the Document Control section represent the classification of this document itself, not the future system, and may therefore differ from the future system's target classification. The H3 authenticated tier is the first time the platform processes per-user personal data โ€” a DPIA is mandatory before Cognito launch (see F6 and FUT-013/FUT-014).


    Future Component S (Spoofing) T (Tampering) R (Repudiation) I (Info Disclosure) D (DoS) E (Elevation) Risk Level
    CIA Data Pipeline Source API spoofing Cached data poisoning Pipeline execution denial Data leakage via cache Pipeline backlog/timeout Pipeline credential escalation HIGH
    Real-Time Voting Dashboard WebSocket connection spoofing Vote data manipulation in transit Connection state denial Vote counting information leak WebSocket flood/connection exhaustion Client-side privilege via WebSocket CRITICAL
    Politician Profile Pages Profile data source spoofing Historical record tampering Profile edit denial Biographical data exposure Profile page DoS via complex queries SEO manipulation for profile ranking MEDIUM
    Automated Translation Pipeline Source language spoofing Translation output manipulation Translation attribution denial Source text leakage Translation queue exhaustion LLM model access escalation HIGH
    Enhanced Dashboards (5 new) Data source spoofing for charts Chart data injection/manipulation Dashboard interaction denial Data aggregation leakage Large dataset rendering DoS Dashboard admin escalation MEDIUM
    EU Parliament Cross-Reference EP MCP Server spoofing Cross-reference data tampering Data linkage denial EU political data leakage API rate limiting/timeout Cross-system privilege escalation MEDIUM
    IMF Data Integration (TypeScript client โ€” scripts/imf-client.ts) IMF origin DNS hijack / TLS MITM IMF JSON response tampering in transit or at rest Stale / mis-vintaged WEO projections cited as current Aggregate public-only; negligible IMF rate-limit (10 req / 5 s) trips workflow Pure-TS client inside the npm SBOM; no new runtime LOW
    ๐ŸŸ  H3 โ€” Bedrock AI Content Engine (Lambda + Step Functions) IAM role assumption / model-endpoint spoofing Indirect prompt injection corrupts generated articles Generation lineage not attributable to a model/vintage Prompt/context leakage via model logs Bedrock throttling / runaway Step-Functions loops Over-scoped Lambda execution role escalates in account HIGH
    ๐ŸŸ  H3 โ€” SageMaker Predictive Models Forged feature inputs / endpoint spoofing Training-data & Feature-Store poisoning skews forecasts Forecast provenance & training-set hash unrecorded Model/feature exposure via misconfigured endpoint Inference-endpoint flooding (pay-per-invoke abuse) Notebook/training-job role escalation HIGH
    ๐ŸŸ  H3 โ€” Neptune Knowledge Graph + Bedrock Knowledge Base Source-document spoofing into the graph/KB Graph-edge tampering; RAG vector poisoning Ingestion source not traceable Embedding inversion / sensitive linkage inference Expensive openCypher/Gremlin or RAG query DoS Cross-tenant graph/KB access via broken IAM HIGH
    ๐ŸŸ  H3 โ€” Aurora Serverless v2 / OpenSearch / Timestream / DynamoDB Lambda resolver spoofs DB identity SQL/NoSQL injection, stored-data tampering DB audit trail gap (no CloudTrail data events) Bulk data exfiltration via broken object-level authZ Query-of-death / connection exhaustion KMS key misuse decrypts at-rest data CRITICAL
    ๐ŸŸ  H3 โ€” AppSync GraphQL + API Gateway Public API Resolver/identity spoofing Mutation tampering, response rewriting Request attribution gap across resolvers Over-fetch / introspection data leakage Deep/complex query & subscription-flood DoS Authorizer bypass elevates to privileged scope HIGH
    ๐ŸŸ  H3 โ€” Amazon Cognito Identity Credential stuffing / token replay Profile & saved-search tampering Disputed account actions (weak audit) IDOR exposes another user's saved data Auth-endpoint flood / token-mint abuse MFA bypass / privilege escalation to admin pool HIGH
    ๐ŸŸ  H3 โ€” Amplify Web PWA + Native Mobile Apps Push-notification / deep-link spoofing Client-side data & cache tampering Device-side action repudiation Insecure local storage / key leakage App-store-targeted client DoS Cert-pinning bypass โ†’ API abuse MEDIUM
    ๐ŸŸ  H3 โ€” Conversational AI (Bedrock Agents, Lex, Transcribe/Polly) Voice/session spoofing Indirect prompt injection via voice/KB context Agent action chain not auditable Briefing leakage across user sessions Agent loop / tool-chain resource exhaustion Excessive agency โ€” agent invokes unintended tools/writes CRITICAL
    ๐ŸŸ  H3 โ€” Multi-Region Resilience (Aurora/DynamoDB Global, S3 CRR, Route 53) Route 53 / health-check spoofing Replication-stream tampering, split-brain writes Cross-region action attribution gap Replica in weaker-controlled region leaks data Failover-triggering DoS, replication lag Regional IAM drift grants stale privileges HIGH
    ๐ŸŸ  H3 โ€” Nordic & EU Federation Data Mesh Foreign-parliament API spoofing Cross-jurisdiction record tampering Federated provenance ambiguity Comparative-dataset linkage disclosure Multi-source fetch amplification DoS Federation trust-boundary privilege crossing MEDIUM
    flowchart TB
    subgraph H2_JEWELS["๐Ÿ”ต H2 Crown Jewels (Static-Deep)"]
    REALTIME["๐Ÿ“Š Real-Time Voting Data<br/>Live parliamentary decisions"]
    POLITICIAN["๐Ÿ‘ค Politician Profiles<br/>Historical performance records"]
    TRANSLATION["๐ŸŒ Translation Pipeline<br/>14-language content integrity"]
    CIA_DATA["๐Ÿ›๏ธ CIA Intelligence Data<br/>19 visualization products"]
    end

    subgraph H3_JEWELS["๐ŸŸ  H3 Crown Jewels (AWS Serverless)"]
    GRAPH["๐Ÿ•ธ๏ธ Neptune Knowledge Graph<br/>+ Bedrock KB vectors"]
    FORECAST["๐Ÿ”ฎ SageMaker Forecast Models<br/>Election & coalition predictions"]
    AURORA["๐Ÿ—„๏ธ Aurora/OpenSearch Stores<br/>Authoritative managed data"]
    IDENTITY["๐Ÿ”‘ Cognito Identities<br/>First authenticated user tier"]
    AGENTS_AI["๐Ÿ’ฌ Bedrock Agents<br/>Autonomous conversational AI"]
    end

    subgraph FUTURE_VECTORS["โš”๏ธ Future Attack Vectors"]
    WEBSOCKET["๐Ÿ”Œ WebSocket Exploitation"]
    CACHE_POISON["๐Ÿ’‰ Cache Poisoning"]
    LLM_MULTI["๐Ÿค– Multi-LLM Orchestration Attack"]
    API_CHAIN["๐Ÿ”— API Chain Compromise"]
    RAG_POISON["๐Ÿ“š RAG / KB Poisoning"]
    IAM_ABUSE["๐Ÿชช Cloud IAM & Token Abuse"]
    EXCESS_AGENCY["๐Ÿง  Agentic Excessive Agency"]
    end

    subgraph FUTURE_AGENTS["๐Ÿ‘ฅ Elevated Threat Agents"]
    ELECTION_ACTOR["๐Ÿ—ณ๏ธ Election Interference Actor"]
    AI_ADVERSARY["๐Ÿค– AI-Enabled Adversary"]
    STATE_ACTOR["๐Ÿ›๏ธ Nation-State APT"]
    CLOUD_ATTACKER["โ˜๏ธ Cloud-Native Attacker"]
    end

    WEBSOCKET --> REALTIME
    CACHE_POISON --> CIA_DATA
    LLM_MULTI --> TRANSLATION
    API_CHAIN --> POLITICIAN
    RAG_POISON --> GRAPH
    RAG_POISON --> FORECAST
    IAM_ABUSE --> AURORA
    IAM_ABUSE --> IDENTITY
    EXCESS_AGENCY --> AGENTS_AI

    ELECTION_ACTOR --> WEBSOCKET
    AI_ADVERSARY --> LLM_MULTI
    AI_ADVERSARY --> EXCESS_AGENCY
    STATE_ACTOR --> CACHE_POISON
    CLOUD_ATTACKER --> IAM_ABUSE
    CLOUD_ATTACKER --> RAG_POISON

    style REALTIME fill:#ffcdd2,stroke:#d32f2f,color:#000
    style POLITICIAN fill:#fff3e0,stroke:#ff9800,color:#000
    style TRANSLATION fill:#e3f2fd,stroke:#2196f3,color:#000
    style CIA_DATA fill:#ffcdd2,stroke:#d32f2f,color:#000
    style GRAPH fill:#ffcdd2,stroke:#d32f2f,color:#000
    style FORECAST fill:#fff3e0,stroke:#ff9800,color:#000
    style AURORA fill:#ffcdd2,stroke:#d32f2f,color:#000
    style IDENTITY fill:#fff3e0,stroke:#ff9800,color:#000
    style AGENTS_AI fill:#ffcdd2,stroke:#d32f2f,color:#000

    Attribute Detail
    Threat Agent Nation-state actor, hacktivist
    Attack Vector WebSocket data injection, man-in-the-middle on data feed
    Target Real-time voting dashboard during live parliamentary vote
    Impact Display incorrect vote counts, undermine democratic trust
    Likelihood Medium (requires intercepting data stream)
    Risk Score 8.5/10 CRITICAL
    MITRE ATT&CK T1565 Data Manipulation, T1557 MITM
    Planned Controls TLS 1.3 for WebSocket, server-side data signing, client-side signature verification, comparison with official riksdagen.se data
    Attribute Detail
    Threat Agent Sophisticated attacker with CIA platform access knowledge
    Attack Vector Compromise cached CIA export data between fetch and display
    Target 19 CIA visualization products cached locally
    Impact Display manipulated political intelligence data across all dashboards
    Likelihood Low (requires pipeline or storage compromise)
    Risk Score 7.2/10 HIGH
    MITRE ATT&CK T1195 Supply Chain Compromise, T1565.001 Stored Data Manipulation
    Planned Controls JSON Schema validation, cryptographic integrity hashing, freshness monitoring (<24h), comparison with source checksums
    Attribute Detail
    Threat Agent AI-enabled adversary, insider threat
    Attack Vector Coordinate prompt injection across multiple AI workflows to create consistent disinformation
    Target News pipeline aggregate+render scripts + multiple per-type workflows (news-evening-analysis, news-realtime-monitor, news-propositions, news-motions, news-committee-reports, news-interpellations, news-week-ahead, news-month-ahead, news-weekly-review, news-monthly-review) consuming the same analysis/daily/$DATE/$SUB/ artifacts
    Impact Consistent AI-generated disinformation across all news outputs, bypassing single-workflow detection
    Likelihood Low (requires deep understanding of multiple workflow prompts)
    Risk Score 7.8/10 HIGH
    MITRE ATT&CK T1659 Content Injection
    Planned Controls Cross-workflow consistency validation, independent fact-checking per workflow, rate limiting on AI content volume, mandatory human review for correlated outputs
    Attribute Detail
    Threat Agent Nation-state actor targeting specific language communities
    Attack Vector Manipulate automated translation to inject politically biased content in specific languages
    Target Arabic, Chinese, or Korean translations (harder for Swedish team to verify)
    Impact Language-specific disinformation targeting diaspora communities
    Likelihood Medium (translation verification is resource-intensive)
    Risk Score 6.8/10 HIGH
    MITRE ATT&CK T1659 Content Injection
    Planned Controls Back-translation verification, native speaker review network, translation consistency scoring, data-translate attribute validation

    The scenarios above (F1โ€“F4) are Horizon 2 threats โ€” they materialise while the platform is still static. The scenarios below (F5โ€“F12) are Horizon 3 threats that only become live once managed AWS services are provisioned; they are pre-modeled so controls ship with each service (FUTURE_ARCHITECTURE.md ยง3, ยง11.4 AWS Security Services).

    Attribute Detail
    Threat Agent Cloud-native attacker, AI-enabled adversary
    Attack Vector Inject crafted documents into the Bedrock Knowledge Base ingestion path so RAG retrieval surfaces poisoned context to Claude during answer generation
    Target Bedrock Knowledge Bases (109K+ document vectors) feeding natural-language queries and conversational AI
    Impact Authoritative-looking but fabricated citations and political analysis served to citizens and journalists
    Likelihood Medium (ingestion pipeline is the soft target, not the model)
    Risk Score 8.0/10 HIGH
    MITRE ATT&CK T1565.001 Stored Data Manipulation, T1195 Supply Chain Compromise
    Planned Controls Signed/whitelisted ingestion sources only (Riksdag/Regeringen/SCB/IMF), embedding-time provenance tags, citation-back-to-dok_id verification, RAG answer grounding score threshold, human review gate for conversational outputs
    Attribute Detail
    Threat Agent Cybercriminal, hacktivist
    Attack Vector Credential stuffing, OAuth token theft/replay, or broken object-level authorization on saved-search/alert resources
    Target Amazon Cognito user pool โ€” the platform's first authenticated tier (saved searches, alerts, personalization)
    Impact Account takeover, exposure of a citizen's political-interest profile (GDPR Art. 9 special category), defacement of personalized content
    Likelihood Medium (authenticated tier is a brand-new attack surface for the platform)
    Risk Score 7.5/10 HIGH
    MITRE ATT&CK T1110 Brute Force, T1539 Steal Web Session Cookie
    Planned Controls Mandatory MFA, Cognito advanced security (adaptive auth + compromised-credential detection), per-user resource-scoped IAM, short-lived tokens + rotation, DPIA before launch, data minimization (no political opinions persisted server-side beyond saved queries)
    Attribute Detail
    Threat Agent Nation-state APT, insider threat
    Attack Vector Over-scoped Lambda execution role or AppSync resolver chained to read Aurora/OpenSearch/DynamoDB beyond its purpose; KMS key misuse to decrypt at rest
    Target Aurora Serverless v2 political_data DB, OpenSearch indices, DynamoDB global tables
    Impact Bulk exfiltration or silent tampering of authoritative political datasets across regions
    Likelihood Low (requires account-level foothold)
    Risk Score 8.5/10 CRITICAL
    MITRE ATT&CK T1078.004 Cloud Accounts, T1530 Data from Cloud Storage, T1213 Data from Information Repositories
    Planned Controls Least-privilege IAM per function (one role per Lambda), VPC isolation + private endpoints, KMS key policies with grant constraints, CloudTrail data events on all stores, GuardDuty + Security Hub correlation, IAM Access Analyzer in CI
    Attribute Detail
    Threat Agent AI-enabled adversary
    Attack Vector Indirect prompt injection (via voice, KB context, or user query) steers a Bedrock Agent to chain tools beyond intent โ€” triggering writes, external calls, or content publication
    Target Conversational AI multi-agent system (Bedrock Agents, Lex, AppSync subscriptions)
    Impact Autonomous publication of manipulated content or unauthorized state changes without human review
    Likelihood Low-Medium (depends on agent tool scope)
    Risk Score 8.2/10 CRITICAL
    MITRE ATT&CK T1659 Content Injection, T1648 Serverless Execution
    Planned Controls Read-only default agent tool scope, write-action approval gates (human-in-the-loop per AI_Policy), tool allowlisting, per-session sandboxing, output-volume limits, full agent-action audit trail
    Attribute Detail
    Threat Agent Election interference actor, nation-state APT
    Attack Vector Poison SageMaker Feature Store / training data or forge inference inputs to skew published seat or coalition forecasts ahead of the 2026 (and later) elections
    Target SageMaker Serverless Inference election/coalition/MP-vote models
    Impact Biased forecasts erode democratic trust and could nudge voter behaviour โ€” a direct attack on neutrality
    Likelihood Medium (high-value target during election windows)
    Risk Score 8.0/10 HIGH
    MITRE ATT&CK T1565 Data Manipulation, T1195.003 Compromise Hardware/Model Supply Chain
    Planned Controls Versioned + hashed training datasets, Feature Store access controls, model-card provenance, published confidence intervals + methodology transparency, cross-validation against SCB/poll aggregates, expert political-scientist review before publication
    Attribute Detail
    Threat Agent Competitor, cybercriminal, hacktivist
    Attack Vector Deeply nested/complex GraphQL queries, schema introspection over-fetch, subscription floods, or API-key/usage-plan abuse on the public REST API
    Target AWS AppSync GraphQL API + Amazon API Gateway public REST endpoints
    Impact Cost-amplification DoS, data scraping at scale, real-time subscription hijacking
    Likelihood Medium (public API is internet-reachable by design)
    Risk Score 6.5/10 MEDIUM
    MITRE ATT&CK T1499 Endpoint DoS, T1190 Exploit Public-Facing Application
    Planned Controls Query depth/complexity limits, disabled production introspection, AWS WAF rate-based + bot-control rules, API Gateway usage plans + key rotation, per-identity throttling, Shield Standard DDoS
    Attribute Detail
    Threat Agent Nation-state APT
    Attack Vector Route 53 / health-check spoofing forces failover to a weaker-controlled region; replication-stream tampering or split-brain writes corrupt Aurora Global / DynamoDB Global tables
    Target Active-passive multi-region deployment (Aurora Global DB, DynamoDB Global Tables, S3 CRR, Route 53)
    Impact Integrity divergence between regions, stale or tampered data served during failover
    Likelihood Low (requires DNS/control-plane compromise)
    Risk Score 6.0/10 MEDIUM
    MITRE ATT&CK T1565.002 Transmitted Data Manipulation, T1583.002 DNS Server
    Planned Controls DNSSEC + Route 53 health-check authentication, consistent cross-region IAM via SCPs, replication integrity checksums, conflict-resolution policy, automated AWS Resilience Hub failover drills, regional config-drift detection (AWS Config)
    Attribute Detail
    Threat Agent Nation-state actor, competitor
    Attack Vector Spoof or tamper a foreign-parliament feed (DK/NO/FI or EU Parliament) so comparative cross-country analysis carries manipulated data through a trusted federation boundary
    Target Shared data-mesh comparative analytics across Nordic & EU parliaments
    Impact Cross-border disinformation laundered through Riksdagsmonitor's neutrality reputation
    Likelihood Low-Medium (each new source widens the trust boundary)
    Risk Score 6.2/10 MEDIUM
    MITRE ATT&CK T1199 Trusted Relationship, T1565.001 Stored Data Manipulation
    Planned Controls Per-source TLS pinning + provenance tagging, source-of-truth precedence rules, cross-source consistency scoring, per-jurisdiction freshness SLAs, explicit federation trust-boundary documentation in FUTURE_DATA_MODEL.md

    Control ID Control Name Future Component STRIDE Coverage Implementation Target Priority
    FUT-001 WebSocket TLS + Data Signing Real-Time Voting Dashboard T, S Q3 2026 ๐Ÿ”ด Critical
    FUT-002 CIA Pipeline JSON Schema Validation CIA Data Pipeline T, I Q2 2026 ๐Ÿ”ด Critical
    FUT-003 Pipeline Cryptographic Integrity CIA Data Pipeline T, R Q2 2026 ๐Ÿ”ด Critical
    FUT-004 Cross-Workflow Consistency Checks AI Content Pipelines T, I Q2 2026 ๐Ÿ”ด Critical
    FUT-005 Back-Translation Verification Translation Pipeline T Q3 2026 ๐ŸŸก High
    FUT-006 Profile Data Source Verification Politician Profiles S, T Q3 2026 ๐ŸŸก High
    FUT-007 Dashboard Data Rate Limiting Enhanced Dashboards D Q2 2026 ๐ŸŸก High
    FUT-008 EU Parliament API Authentication EU Cross-Reference S, E Q4 2026 ๐ŸŸก High
    FUT-009 Real-Time Anomaly Detection Real-Time Dashboard T, I Q3 2026 ๐Ÿ”ด Critical
    FUT-010 Automated Content Volume Limiting AI Workflows D, T Q2 2026 ๐ŸŸก High
    FUT-011 RAG Source Allowlist + Provenance Tagging Bedrock Knowledge Base T, I 2027 Q2 ๐Ÿ”ด Critical
    FUT-012 RAG Grounding-Score Threshold + Citation Verification Bedrock KB / Conversational AI T 2027 Q2 ๐Ÿ”ด Critical
    FUT-013 Cognito MFA + Advanced Security (adaptive auth) Cognito Identity S, E 2027 Q4 ๐Ÿ”ด Critical
    FUT-014 Per-User Resource-Scoped Authorization (anti-IDOR) Cognito / AppSync / Aurora I, E 2027 Q4 ๐Ÿ”ด Critical
    FUT-015 Least-Privilege IAM per Lambda + Access Analyzer in CI Lambda / IAM E 2027 ๐Ÿ”ด Critical
    FUT-016 CloudTrail Data Events + GuardDuty/Security Hub Correlation All managed stores R, I 2027 ๐ŸŸก High
    FUT-017 Agent Tool Allowlist + Write-Action Approval Gate Bedrock Agents E, T 2028 ๐Ÿ”ด Critical
    FUT-018 Versioned/Hashed Training Data + Feature-Store Access Control SageMaker T 2026 Q4 ๐ŸŸก High
    FUT-019 GraphQL Depth/Complexity Limits + WAF Rate Rules AppSync / API Gateway D 2027 ๐ŸŸก High
    FUT-020 KMS Key Policies + Envelope Encryption (at rest) Aurora / DynamoDB / S3 I 2027 ๐Ÿ”ด Critical
    FUT-021 DNSSEC + Cross-Region SCP/Config-Drift Detection Multi-Region Resilience T, E 2028 ๐ŸŸก High
    FUT-022 Federation Trust-Boundary Provenance + Consistency Scoring Nordic/EU Data Mesh S, T 2027โ€“2030 ๐ŸŸก High
    STRIDE Category Future Primary Control Future Secondary Control Future Monitoring
    Spoofing WebSocket TLS (FUT-001), API auth (FUT-008), Cognito MFA (FUT-013) Data source verification (FUT-006), federation provenance (FUT-022) Connection/auth logs, GuardDuty (FUT-016)
    Tampering JSON Schema validation (FUT-002), data signing (FUT-003), RAG allowlist (FUT-011) Cross-workflow checks (FUT-004), training-data hashing (FUT-018) Data integrity monitoring, CloudTrail data events (FUT-016)
    Repudiation Cryptographic integrity (FUT-003), CloudTrail data events (FUT-016) Git-based change tracking, agent-action audit (FUT-017) Audit trail analysis, Security Hub
    Info Disclosure Resource-scoped authZ (FUT-014), KMS at rest (FUT-020) Rate limiting (FUT-007), RAG grounding (FUT-012) Data access monitoring, Access Analyzer (FUT-015)
    DoS Rate limiting (FUT-007), GraphQL complexity limits (FUT-019) WebSocket/connection limits, WAF + Shield Performance monitoring, anomaly detection (FUT-009)
    Elevation Least-privilege IAM per Lambda (FUT-015), Cognito MFA (FUT-013) Agent write-approval gate (FUT-017), cross-region SCP (FUT-021) Privilege usage monitoring, IAM Access Analyzer

    Threat Agent Motivation Capability Future Target Risk Trend
    Nation-State APT Political influence, intelligence gathering Very High (zero-day, AI-enhanced) Real-time voting data, politician profiles โฌ†๏ธ Increasing
    AI-Enabled Adversary Automated exploitation, disinformation High (LLM-driven attacks) Translation pipeline, multi-workflow orchestration โฌ†๏ธ Rapidly increasing
    Hacktivist Political disruption, ideology Medium (commodity tools + AI) Public dashboards, election forecasts โžก๏ธ Stable
    Insider Threat Data manipulation, sabotage High (pipeline access) CIA data pipeline, content generation โฌ†๏ธ Increasing with more contributors
    Competitor Market intelligence, replication Medium (OSINT, scraping) Dashboard algorithms, analysis methodology โžก๏ธ Stable
    Cybercriminal Ransomware, cryptomining Medium (supply chain focus) CI/CD pipeline, dependency chain โฌ†๏ธ Increasing
    Cloud-Native Attacker (H3) Account compromise, data theft, cost-amplification High (IAM abuse, serverless/RAG exploitation) Aurora/OpenSearch stores, Cognito identities, Bedrock KB, public AppSync/API Gateway โฌ†๏ธ Emerging with AWS migration
    flowchart TD
    ROOT["๐ŸŽฏ Manipulate Real-Time Vote Display"]

    A1["โš”๏ธ A1: Compromise Data Feed<br/>Probability: 15%"]
    A2["โš”๏ธ A2: Man-in-the-Middle<br/>Probability: 10%"]
    A3["โš”๏ธ A3: Client-Side Injection<br/>Probability: 20%"]
    A4["โš”๏ธ A4: Cache Poisoning<br/>Probability: 12%"]

    A1_1["๐Ÿ”ง A1.1: Compromise Riksdag API proxy"]
    A1_2["๐Ÿ”ง A1.2: DNS hijacking of data source"]
    A2_1["๐Ÿ”ง A2.1: TLS downgrade attack"]
    A2_2["๐Ÿ”ง A2.2: WebSocket hijacking"]
    A3_1["๐Ÿ”ง A3.1: XSS via dashboard parameter"]
    A3_2["๐Ÿ”ง A3.2: Browser extension manipulation"]
    A4_1["๐Ÿ”ง A4.1: CDN cache poisoning"]
    A4_2["๐Ÿ”ง A4.2: LocalStorage corruption"]

    ROOT --> A1
    ROOT --> A2
    ROOT --> A3
    ROOT --> A4
    A1 --> A1_1
    A1 --> A1_2
    A2 --> A2_1
    A2 --> A2_2
    A3 --> A3_1
    A3 --> A3_2
    A4 --> A4_1
    A4 --> A4_2

    style ROOT fill:#ffcdd2,stroke:#d32f2f,color:#000
    style A1 fill:#fff3e0,stroke:#ff9800,color:#000
    style A2 fill:#fff3e0,stroke:#ff9800,color:#000
    style A3 fill:#e3f2fd,stroke:#2196f3,color:#000
    style A4 fill:#fff3e0,stroke:#ff9800,color:#000
    flowchart TD
    ROOT2["๐ŸŽฏ Compromise CIA Intelligence Data"]

    B1["โš”๏ธ B1: Pipeline Source Compromise<br/>Probability: 8%"]
    B2["โš”๏ธ B2: Cache/Storage Manipulation<br/>Probability: 12%"]
    B3["โš”๏ธ B3: Schema Bypass<br/>Probability: 5%"]
    B4["โš”๏ธ B4: AI Content Poisoning<br/>Probability: 15%"]

    B1_1["๐Ÿ”ง B1.1: CIA platform API manipulation"]
    B1_2["๐Ÿ”ง B1.2: Nightly fetch interception"]
    B2_1["๐Ÿ”ง B2.1: GitHub CDN cache poisoning"]
    B2_2["๐Ÿ”ง B2.2: S3 bucket manipulation"]
    B3_1["๐Ÿ”ง B3.1: Schema version mismatch exploit"]
    B3_2["๐Ÿ”ง B3.2: JSON Schema validation bypass"]
    B4_1["๐Ÿ”ง B4.1: Prompt injection via CIA data fields"]
    B4_2["๐Ÿ”ง B4.2: Cross-workflow data flow contamination"]

    ROOT2 --> B1
    ROOT2 --> B2
    ROOT2 --> B3
    ROOT2 --> B4
    B1 --> B1_1
    B1 --> B1_2
    B2 --> B2_1
    B2 --> B2_2
    B3 --> B3_1
    B3 --> B3_2
    B4 --> B4_1
    B4 --> B4_2

    style ROOT2 fill:#ffcdd2,stroke:#d32f2f,color:#000
    style B1 fill:#fff3e0,stroke:#ff9800,color:#000
    style B2 fill:#fff3e0,stroke:#ff9800,color:#000
    style B3 fill:#e8f5e9,stroke:#4caf50,color:#000
    style B4 fill:#fff3e0,stroke:#ff9800,color:#000
    flowchart TD
    ROOT3["๐ŸŽฏ Exfiltrate / Tamper Authoritative AWS Data Stores"]

    C1["โš”๏ธ C1: Identity Compromise<br/>Probability: 10%"]
    C2["โš”๏ธ C2: Over-Scoped Role Abuse<br/>Probability: 9%"]
    C3["โš”๏ธ C3: RAG / KB Poisoning<br/>Probability: 14%"]
    C4["โš”๏ธ C4: Public API Abuse<br/>Probability: 18%"]

    C1_1["๐Ÿ”ง C1.1: Cognito credential stuffing / token replay"]
    C1_2["๐Ÿ”ง C1.2: CI/CD OIDC role assumption"]
    C2_1["๐Ÿ”ง C2.1: Lambda execution-role privilege escalation"]
    C2_2["๐Ÿ”ง C2.2: KMS key-policy misuse to decrypt at rest"]
    C3_1["๐Ÿ”ง C3.1: Poison Bedrock KB ingestion source"]
    C3_2["๐Ÿ”ง C3.2: SageMaker feature-store / training poisoning"]
    C4_1["๐Ÿ”ง C4.1: GraphQL depth/complexity cost-DoS"]
    C4_2["๐Ÿ”ง C4.2: IDOR on AppSync resolver / saved data"]

    ROOT3 --> C1
    ROOT3 --> C2
    ROOT3 --> C3
    ROOT3 --> C4
    C1 --> C1_1
    C1 --> C1_2
    C2 --> C2_1
    C2 --> C2_2
    C3 --> C3_1
    C3 --> C3_2
    C4 --> C4_1
    C4 --> C4_2

    style ROOT3 fill:#ffcdd2,stroke:#d32f2f,color:#000
    style C1 fill:#fff3e0,stroke:#ff9800,color:#000
    style C2 fill:#ffcdd2,stroke:#d32f2f,color:#000
    style C3 fill:#fff3e0,stroke:#ff9800,color:#000
    style C4 fill:#e3f2fd,stroke:#2196f3,color:#000
    Kill Chain Phase Future Attack Capability Disruption Control Detection Mechanism
    Reconnaissance AI-powered API enumeration of new endpoints Rate limiting, API key rotation (FUT-008) API access pattern monitoring
    Weaponization LLM-crafted prompt injection payloads Input validation, prompt sanitization (FUT-004) Prompt content analysis logs
    Delivery Compromised data in CIA pipeline/WebSocket feeds TLS 1.3 pinning, source verification (FUT-001, FUT-002) Network traffic anomaly detection
    Exploitation Schema bypass, translation model manipulation JSON Schema strict validation (FUT-002), model input filtering Validation failure alerts, output consistency checking
    Installation Persistent cache poisoning, LocalStorage manipulation Cache TTL enforcement, integrity hashing (FUT-003) Cache integrity monitoring
    C2 AI-orchestrated multi-workflow coordination Cross-workflow consistency checks (FUT-004), volume limiting (FUT-010) Workflow correlation analysis
    Actions on Objectives Public disinformation via manipulated dashboards/news Human review gate, source cross-validation, fact-checking Content integrity alerts, user reporting

    Future Feature New Endpoints Data Sensitivity External Dependencies Attack Surface Rating
    Real-Time Voting Dashboard WebSocket endpoint, SSE stream Critical (live democratic data) Riksdag API, CDN ๐Ÿ”ด High
    CIA Data Pipeline Nightly fetch endpoint, cache API High (19 intelligence products) CIA Platform API, S3 ๐Ÿ”ด High
    Politician Profile Pages Per-MP URL routes (349+ pages) High (career/voting history) CIA data, Riksdag API ๐ŸŸก Medium
    Automated Translation LLM API calls (14 languages) Medium (content integrity) LLM Provider API ๐ŸŸก Medium
    EU Parliament Cross-Ref EP MCP Server API, GraphQL Medium (EU political data) EP Open Data API ๐ŸŸข Low
    5 New Dashboards Chart data endpoints, D3 renders Medium (aggregated analytics) CIA data, Chart.js CDN ๐ŸŸก Medium
    ๐ŸŸ  H3 โ€” Bedrock AI Content Engine Lambda invoke, Bedrock/Polly model endpoints, Step Functions High (generated public content integrity) Amazon Bedrock, Polly ๐Ÿ”ด High
    ๐ŸŸ  H3 โ€” Neptune + Bedrock Knowledge Base openCypher/Gremlin, RAG retrieve/query High (semantic intelligence) Neptune Serverless, Bedrock KB ๐Ÿ”ด High
    ๐ŸŸ  H3 โ€” Aurora/OpenSearch/Timestream/DynamoDB Lambda DB resolvers (private) Critical (authoritative data) AWS managed data services, KMS ๐Ÿ”ด High
    ๐ŸŸ  H3 โ€” AppSync GraphQL + API Gateway Public GraphQL + REST endpoints, subscriptions High (public API, scraping/DoS target) AppSync, API Gateway, WAF ๐Ÿ”ด High
    ๐ŸŸ  H3 โ€” Cognito Identity Auth/token endpoints, user-pool APIs High (Art. 9 user profiles) Amazon Cognito ๐Ÿ”ด High
    ๐ŸŸ  H3 โ€” SageMaker Predictive Models Serverless inference endpoints High (forecast integrity) SageMaker, Feature Store ๐ŸŸก Medium
    ๐ŸŸ  H3 โ€” Amplify Web PWA + Mobile Apps App API calls, push, deep links Medium (client integrity) Amplify, CloudFront, Shield ๐ŸŸก Medium
    ๐ŸŸ  H3 โ€” Conversational AI (Agents/Lex) Chat/voice sessions, agent tool calls Critical (autonomous actions) Bedrock Agents, Lex, Transcribe ๐Ÿ”ด High
    ๐ŸŸ  H3 โ€” Multi-Region Resilience Route 53 failover, cross-region replication High (integrity across regions) Aurora/DynamoDB Global, S3 CRR ๐ŸŸก Medium
    ๐ŸŸ  H3 โ€” Nordic/EU Federation Foreign-parliament + EP API ingestion Medium (cross-jurisdiction integrity) DK/NO/FI + EU Parliament APIs ๐ŸŸก Medium
    flowchart LR
    subgraph EXTERNAL["๐ŸŒ External Sources"]
    RIKSDAG_API["Riksdag API"]
    CIA_API["CIA Platform"]
    EP_API["EU Parliament API"]
    LLM_API["LLM Provider"]
    end

    subgraph PIPELINE["โš™๏ธ Data Pipeline"]
    FETCH["Nightly Fetch"]
    VALIDATE["Schema Validation"]
    TRANSFORM["Data Transform"]
    CACHE["Cache Layer"]
    end

    subgraph DELIVERY["๐Ÿ“ฆ Content Delivery"]
    CDN["CloudFront CDN"]
    S3["S3 Static Assets"]
    PAGES["GitHub Pages"]
    end

    subgraph CLIENT["๐Ÿ–ฅ๏ธ Browser Client"]
    DASHBOARD["Interactive Dashboards"]
    REALTIME["Real-Time Feeds"]
    PROFILES["Politician Profiles"]
    end

    RIKSDAG_API -->|"๐Ÿ”ด T: Data interception"| FETCH
    CIA_API -->|"๐Ÿ”ด T: Source compromise"| FETCH
    EP_API -->|"๐ŸŸก S: API spoofing"| FETCH
    LLM_API -->|"๐Ÿ”ด T: Response manipulation"| TRANSFORM

    FETCH -->|"๐ŸŸก T: Pipeline tampering"| VALIDATE
    VALIDATE -->|"๐ŸŸก I: Validation bypass"| TRANSFORM
    TRANSFORM -->|"๐ŸŸก T: Cache poisoning"| CACHE

    CACHE -->|"๐ŸŸก T: CDN poisoning"| CDN
    CACHE --> S3
    CACHE --> PAGES

    CDN --> DASHBOARD
    CDN --> REALTIME
    CDN --> PROFILES

    style EXTERNAL fill:#e3f2fd,stroke:#2196f3,color:#000
    style PIPELINE fill:#fff3e0,stroke:#ff9800,color:#000
    style DELIVERY fill:#e8f5e9,stroke:#4caf50,color:#000
    style CLIENT fill:#f3e5f5,stroke:#9c27b0,color:#000

    OWASP LLM ID Threat Future Relevance Planned Mitigation
    LLM01 Prompt Injection ๐Ÿ”ด Critical โ€” More workflows = larger injection surface Per-workflow input sanitization, prompt boundary enforcement
    LLM02 Insecure Output Handling ๐Ÿ”ด Critical โ€” Auto-generated content directly published HTML sanitization, output schema validation, human review gate
    LLM03 Training Data Poisoning ๐ŸŸก Medium โ€” Indirect via MCP data sources Source integrity verification, data provenance tracking
    LLM04 Model Denial of Service ๐ŸŸก Medium โ€” Multiple concurrent workflow runs Workflow concurrency limits, timeout enforcement, rate limiting
    LLM05 Supply Chain Vulnerabilities ๐ŸŸก Medium โ€” LLM model updates may introduce regressions Model version pinning, output regression testing
    LLM06 Sensitive Information Disclosure ๐ŸŸข Low โ€” Public data only, no PII Data classification enforcement, output filtering
    LLM07 Insecure Plugin Design ๐Ÿ”ด Critical โ€” MCP server tools are "plugins" MCP tool allowlisting, capability-based access control
    LLM08 Excessive Agency ๐Ÿ”ด Critical โ€” Agents can create/edit content + trigger workflows Write operation approval gates, output volume limits
    LLM09 Overreliance ๐ŸŸก Medium โ€” Over-trusting AI-generated political analysis Mandatory human editorial review, confidence scoring
    LLM10 Model Theft ๐ŸŸข Low โ€” Using commercial API, not custom model API key rotation, access logging

    Mapping note: the table above uses the OWASP LLM Top-10 (2023/2024) IDs already established in this document. For Horizon 3 the same risks intensify as the platform moves from build-time MCP agents to managed Bedrock Agents, Knowledge Bases (RAG) and SageMaker models. The H3-specific intensification is summarised below.

    OWASP LLM Risk H3 Driver H3-Specific Mitigation
    LLM01 Prompt Injection Indirect injection via RAG KB context, voice (Transcribe), and user queries to Bedrock Agents Source allowlist (FUT-011), grounding-score threshold (FUT-012), per-session sandboxing
    LLM02 Insecure Output Handling Agents can publish content autonomously to S3/CloudFront Write-action approval gate (FUT-017), output schema validation, human review gate
    LLM03/04 Data Poisoning & RAG Manipulation Bedrock KB ingestion + SageMaker Feature Store are poisonable RAG provenance tagging (FUT-011), versioned/hashed training data (FUT-018)
    LLM06 Sensitive Information Disclosure Cognito introduces real user profiles (Art. 9) into prompts/briefings Per-user scoping (FUT-014), data minimization, briefing isolation per session
    LLM07 Insecure Plugin Design Bedrock Agent "tools" replace MCP tools as the plugin surface Tool allowlist (FUT-017), capability-scoped IAM (FUT-015)
    LLM08 Excessive Agency Multi-agent autonomous task execution (Phase 4) Read-only default scope, write approval gates (FUT-017), agent-action audit (FUT-016)
    LLM10 Model Theft Custom SageMaker forecasting models now exist Endpoint authZ, model-artifact encryption (FUT-020), access logging
    Workflow Combination Attack Scenario Impact Detection Difficulty Planned Control
    article-generator + evening-analysis Coordinated disinformation: article + supporting analysis Critical Hard โ€” requires cross-workflow correlation FUT-004: Cross-workflow consistency
    translate + article-generator Inject bias in translation of generated content High Hard โ€” translation errors look like hallucinations FUT-005: Back-translation verification
    realtime-monitor + committee-reports Time-sensitive misinformation during live events Critical Medium โ€” timing anomalies detectable FUT-009: Real-time anomaly detection
    propositions + motions + weekly-review Long-running narrative manipulation across weekly content High Very Hard โ€” gradual drift is subtle Longitudinal content consistency analysis
    (H3) Bedrock Agent + Knowledge Base + Step Functions Poisoned KB context drives an agent to autonomously publish manipulated briefings Critical Very Hard โ€” looks like a grounded answer FUT-011, FUT-012, FUT-017
    (H3) SageMaker forecast + news-pre-election workflow Skewed forecast amplified into election-window articles Critical Hard โ€” forecast looks statistically plausible FUT-018 + SCB/poll cross-validation + expert review

    Fielding the Political-Intelligence Capability Catalog (C1โ€“C32) creates a new, high-value attack surface: an adversary who can corrupt the intelligence pipeline can launder a manipulated judgment through the platform's own credibility. These threats are distinct from generic web threats โ€” they target analytic integrity, calibration, neutrality and provenance. The catalog's assurance pillar (C26โ€“C32) exists specifically to counter them.

    Component Threat (STRIDE) Scenario Counter-capability
    Multi-INT fusion graph (C6) Tampering Poisoned edge fabricates a personโ†”funding link C8 evidence anchor (no edge without graded dok_id); human-review hold
    Entity resolution (C1) Spoofing Adversary games identifiers to merge/split entities Deterministic-key + embedding agreement; confidence floor; audit log
    I&W tripwires (C14) Denial of warning Flood of decoy signals desensitizes thresholds / hides real event Adaptive thresholds, anomaly-on-anomaly, human triage gate
    Forecasting + calibration (C13/C29) Tampering / Repudiation Skewed training data degrades Brier; later denial of bias Immutable calibration ledger; rolling Brier as release gate; assumption logs
    FIMI early-warning (C20) Information disclosure / abuse Mission-creep toward citizen profiling; false attribution Hard ethics gate, aggregate-only, advisory-not-accusatory, no profiling
    SAT / estimative engine (C11/C22) Tampering Prompt-injection steers ACH toward a predetermined conclusion C26 injection screening; devil's-advocate pass; ICD-203 + human sign-off
    Provenance / C2PA (C8/C9) Spoofing Forged content credential passes synthetic evidence as authentic KMS-signed manifests; deepfake detector; refuse-to-cite on failure
    Neutrality gate (C31) Elevation / bias injection Asymmetric output ships, eroding party-neutrality CI party-symmetry audit; block-on-asymmetry; dual-control override
    ID Scenario Impact Detection Planned control
    PI-T1 Analytic-pipeline data poisoning โ€” adversary seeds public-looking sources to bias fusion/forecasting Critical โ€” manipulated judgments gain platform credibility Hard โ€” inputs look legitimate Source-grading floor, provenance, outlier detection on ingest, calibration drift alarms
    PI-T2 Prompt-injection of the SAT/estimative agent via crafted document text Critical โ€” steered "reasoned" conclusion Hard โ€” looks like grounded analysis C26 injection screening, Bedrock Guardrails, tool-permission minimization, human sign-off
    PI-T3 Provenance forgery / deepfake evidence cited in a briefing High โ€” false evidence in the record Medium โ€” manifest + detector checks C2PA verification, KMS signing, synthetic-media detector, refuse-to-cite
    PI-T4 Neutrality subversion โ€” gradual asymmetric framing across products Critical โ€” destroys institutional trust Very Hard โ€” gradual drift C31 party-symmetry CI gate, longitudinal symmetry monitoring, dual review
    PI-T5 Warning suppression / decoy flooding of I&W tripwires High โ€” real coalition/vote event missed Medium โ€” signal-rate anomalies Adaptive thresholds, redundancy across indicators, human-on-the-loop
    PI-T6 Calibration gaming โ€” manipulate which questions resolve to inflate apparent accuracy High โ€” misleading trust signal Hard โ€” statistically subtle Pre-registered questions, immutable ledger, independent resolution criteria
    PI-T7 FIMI targeting the platform itself โ€” adversary narratives crafted to trigger false advisories High โ€” platform amplifies adversary frame Hard โ€” designed to look organic Attribution-confidence floors, ethics gate, human framing, advisory-only output
    Scenario STRIDE MITRE ATT&CK / ATLAS OWASP LLM Top 10
    PI-T1 poisoning Tampering ATLAS: ML Supply-Chain / Data Poisoning LLM03 Training-Data Poisoning
    PI-T2 injection Tampering / EoP ATLAS: LLM Prompt Injection LLM01 Prompt Injection
    PI-T3 provenance forgery Spoofing T1565 Data Manipulation LLM08 Excessive Agency (citation)
    PI-T4 neutrality Repudiation / bias โ€” (governance) LLM09 Overreliance
    PI-T5 warning suppression DoS T1499 Endpoint DoS (signal) LLM04 Model DoS
    PI-T6 calibration gaming Repudiation โ€” (integrity) LLM09 Overreliance
    PI-T7 FIMI targeting Information abuse DISARM TTPs LLM09 Overreliance

    Governing principle. Every intelligence-capability threat is met by an integrity-by-construction control, not by trust in the model: evidence anchoring, immutable calibration, provenance signing, neutrality-as-a-CI-gate, and a mandatory human-on-the-loop before any estimative product is published. See FUTURE_SECURITY_ARCHITECTURE.md for the corresponding controls.


    The platform's mission is democratic transparency โ€” any threat that subverts, distorts, or undermines public accountability is existential regardless of technical sophistication.

    Riksdagsmonitor occupies a unique position: a neutral, AI-powered democratic-intelligence platform whose outputs influence citizen understanding of parliamentary proceedings. This creates a category of threats distinct from generic cybersecurity โ€” threats to democratic processes, institutional trust, and political neutrality that no standard web-security framework adequately covers.

    flowchart TB
    subgraph DEMOCRATIC_THREATS["๐Ÿ—ณ๏ธ Democratic Integrity Threats"]
    direction TB
    DT1["๐ŸŽญ Neutrality Subversion<br/>Asymmetric framing across parties"]
    DT2["๐Ÿ“Š Forecast Weaponization<br/>Biased predictions influence voters"]
    DT3["๐Ÿ—ž๏ธ Information Laundering<br/>Adversary narratives gain platform credibility"]
    DT4["๐Ÿ”‡ Accountability Suppression<br/>Hide/downplay political misconduct"]
    DT5["โšก Election-Window Exploitation<br/>Time-critical attacks during campaigns"]
    DT6["๐ŸŒ Cross-Border Influence<br/>Foreign interference via federation"]
    end

    subgraph DEMOCRATIC_CONTROLS["๐Ÿ›ก๏ธ Democratic Safeguards"]
    direction TB
    DC1["โš–๏ธ Party-Symmetry CI Gate<br/>Automated neutrality enforcement"]
    DC2["๐Ÿ“ Calibration Ledger<br/>Immutable forecast accuracy tracking"]
    DC3["๐Ÿ” Source-Grade Floor<br/>Minimum evidence threshold"]
    DC4["๐Ÿ‘๏ธ Human-on-the-Loop<br/>Mandatory editorial oversight"]
    DC5["๐Ÿšซ Election Cooling Period<br/>Restricted AI during election silence"]
    DC6["๐Ÿค Federation Trust Boundaries<br/>Per-source integrity verification"]
    end

    subgraph DEMOCRATIC_ACTORS["๐Ÿ‘ฅ Democratic Threat Actors"]
    direction TB
    DA1["๐Ÿ›๏ธ State-Sponsored IO<br/>Foreign influence operations"]
    DA2["๐ŸŽช Domestic Political Operatives<br/>Partisan manipulation attempts"]
    DA3["๐Ÿค– Autonomous AI Agents<br/>Unintended bias amplification"]
    DA4["๐Ÿ“ฐ Disinformation Networks<br/>Coordinated inauthentic behavior"]
    end

    DA1 --> DT3
    DA1 --> DT6
    DA2 --> DT1
    DA2 --> DT4
    DA3 --> DT1
    DA3 --> DT2
    DA4 --> DT3
    DA4 --> DT5

    DT1 -.->|mitigated by| DC1
    DT2 -.->|mitigated by| DC2
    DT3 -.->|mitigated by| DC3
    DT4 -.->|mitigated by| DC4
    DT5 -.->|mitigated by| DC5
    DT6 -.->|mitigated by| DC6

    style DEMOCRATIC_THREATS fill:#fff3e0,stroke:#e65100,color:#000
    style DEMOCRATIC_CONTROLS fill:#e8f5e9,stroke:#2e7d32,color:#000
    style DEMOCRATIC_ACTORS fill:#fce4ec,stroke:#c62828,color:#000
    Attribute Detail
    ๐ŸŽญ Threat Agent Autonomous AI drift (unintentional), sophisticated insider, domestic political operative
    โš”๏ธ Attack Vector Subtle, consistent asymmetry in AI-generated content: tone, coverage depth, or framing favors one bloc over another across hundreds of articles over weeks/months
    ๐ŸŽฏ Target The platform's core neutrality invariant โ€” equal treatment of all 8 Riksdag parties
    ๐Ÿ’ฅ Impact Institutional credibility destroyed; platform becomes a perceived partisan tool; cited in political campaigns as evidence of bias
    ๐Ÿ“Š Likelihood Medium-High (LLM training biases are well-documented; drift is natural without active correction)
    โš ๏ธ Risk Score 9.0/10 CRITICAL
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1659 Content Injection (adapted: content bias injection)
    ๐Ÿ›ก๏ธ Planned Controls FUT-023: Party-symmetry CI gate (automated), FUT-024: longitudinal sentiment-balance monitoring, dual-review for cross-party articles, mandatory bloc-parity metrics in every weekly review
    Attribute Detail
    ๐ŸŽญ Threat Agent Election interference actor, nation-state information operation
    โš”๏ธ Attack Vector Timing-aware attack: manipulate SageMaker forecast inputs or translation pipeline during the 30-day pre-election window when media amplification is maximal
    ๐ŸŽฏ Target Published seat/coalition predictions, pre-election news coverage, voter information pages
    ๐Ÿ’ฅ Impact Biased forecasts amplified by media; potential violation of Swedish election silence conventions; voter behavior influence; legal/regulatory consequences
    ๐Ÿ“Š Likelihood Medium (high-value target with clear temporal window)
    โš ๏ธ Risk Score 8.8/10 CRITICAL
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1565 Data Manipulation, T1583.006 Web Services
    ๐Ÿ›ก๏ธ Planned Controls FUT-025: Election cooling-period protocol (restricted AI autonomy, mandatory human approval for all election-relevant content), elevated monitoring, cross-validation with SCB/Valmyndigheten, explicit uncertainty disclosure
    Attribute Detail
    ๐ŸŽญ Threat Agent Domestic political operative, insider threat, sophisticated lobbyist
    โš”๏ธ Attack Vector Manipulate content pipeline to suppress, delay, or downplay politically inconvenient information (votes, motions, committee decisions) while amplifying favorable narratives
    ๐ŸŽฏ Target News article generation, politician profile pages, voting record displays
    ๐Ÿ’ฅ Impact Platform becomes complicit in accountability evasion; undermines democratic oversight function; erosion of public trust
    ๐Ÿ“Š Likelihood Low-Medium (requires insider access or pipeline compromise)
    โš ๏ธ Risk Score 7.5/10 HIGH
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1565.001 Stored Data Manipulation, T1070 Indicator Removal
    ๐Ÿ›ก๏ธ Planned Controls FUT-026: Completeness audit (automated check that all Riksdag decisions/votes are covered), source-of-record reconciliation with riksdagen.se, time-to-publish SLA monitoring, dual-control on content deletion
    Attribute Detail
    ๐ŸŽญ Threat Agent Foreign information operation (FIMI), coordinated inauthentic network
    โš”๏ธ Attack Vector Seed manipulated data into upstream sources (Riksdag API responses, government press releases via g0v.se, foreign parliament feeds) knowing Riksdagsmonitor will automatically ingest, validate, and republish โ€” laundering disinformation through the platform's trusted reputation
    ๐ŸŽฏ Target External data ingestion paths: Riksdag API, Regeringen/g0v.se, SCB, IMF, Nordic/EU parliament feeds
    ๐Ÿ’ฅ Impact Platform amplifies state-sponsored disinformation with the credibility of "independently verified" parliamentary analysis
    ๐Ÿ“Š Likelihood Low-Medium (requires compromising or spoofing upstream government sources)
    โš ๏ธ Risk Score 8.0/10 HIGH
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1199 Trusted Relationship, T1659 Content Injection
    ๐Ÿ›ก๏ธ Planned Controls FUT-027: Multi-source cross-validation (never rely on single source), anomaly detection on ingest deltas, provenance chain verification, source-grading with confidence floors, human escalation for statistically improbable data changes

    Horizon 3 introduces the platform's first authenticated user tier โ€” transforming privacy from a non-concern to a critical obligation.

    flowchart LR
    subgraph USER_DATA["๐Ÿ‘ค H3 User Data at Risk"]
    UD1["๐Ÿ” Saved Searches<br/>Reveal political interests"]
    UD2["๐Ÿ”” Alert Subscriptions<br/>Track political monitoring"]
    UD3["๐Ÿ“Š Personalization<br/>Behavioral profile"]
    UD4["๐Ÿ’ฌ Chat History<br/>Political questions asked"]
    end

    subgraph GDPR_THREATS["โš ๏ธ Privacy Threats"]
    GT1["๐Ÿ“‹ Art. 9 Violation<br/>Special-category data exposure"]
    GT2["๐Ÿ•ต๏ธ Profiling Risk<br/>Political opinion inference"]
    GT3["๐ŸŒ Cross-Border Transfer<br/>Multi-region data residency"]
    GT4["๐Ÿ—‘๏ธ Erasure Complexity<br/>Right-to-be-forgotten across replicas"]
    GT5["๐Ÿ”— Linkage Attack<br/>De-anonymization via query patterns"]
    end

    subgraph PRIVACY_CONTROLS["๐Ÿ›ก๏ธ Privacy Controls"]
    PC1["๐Ÿ“ DPIA Mandatory<br/>Before Cognito launch"]
    PC2["๐Ÿ”’ Data Minimization<br/>No opinions stored server-side"]
    PC3["๐Ÿ  EU Data Residency<br/>eu-west-1 primary"]
    PC4["โฑ๏ธ Auto-Purge<br/>Configurable retention"]
    PC5["๐ŸŽญ Pseudonymization<br/>Query-level privacy"]
    end

    UD1 --> GT1
    UD2 --> GT2
    UD3 --> GT5
    UD4 --> GT1

    GT1 -.->|mitigated by| PC1
    GT2 -.->|mitigated by| PC2
    GT3 -.->|mitigated by| PC3
    GT4 -.->|mitigated by| PC4
    GT5 -.->|mitigated by| PC5

    style USER_DATA fill:#e3f2fd,stroke:#1565c0,color:#000
    style GDPR_THREATS fill:#ffebee,stroke:#c62828,color:#000
    style PRIVACY_CONTROLS fill:#e8f5e9,stroke:#2e7d32,color:#000
    Attribute Detail
    ๐ŸŽญ Threat Agent Data breach attacker, insider, law enforcement overreach
    โš”๏ธ Attack Vector Aggregate saved searches, alert patterns, and chatbot questions to infer a user's political opinions โ€” GDPR Article 9 special-category data โ€” without explicit consent for that processing purpose
    ๐ŸŽฏ Target Cognito user profiles + associated DynamoDB/Aurora query history
    ๐Ÿ’ฅ Impact Violation of GDPR Art. 9 (processing special-category data without lawful basis); regulatory fines up to 4% annual turnover; chilling effect on civic engagement
    ๐Ÿ“Š Likelihood Medium (inference is technically straightforward once data is collected)
    โš ๏ธ Risk Score 8.5/10 CRITICAL
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1530 Data from Cloud Storage, T1213 Data from Information Repositories
    ๐Ÿ›ก๏ธ Planned Controls FUT-028: Privacy-by-design architecture (no server-side political-opinion storage), client-side encryption for saved queries, aggregate-only analytics, automated data minimization, DPIA gate before any new data collection, privacy-preserving personalization (on-device ML)
    Attribute Detail
    ๐ŸŽญ Threat Agent Configuration error, multi-region replication misconfiguration
    โš”๏ธ Attack Vector DynamoDB Global Tables or Aurora Global replication copies EU citizen data to non-adequate jurisdictions (e.g., us-east-1) without proper safeguards
    ๐ŸŽฏ Target User personal data in DynamoDB/Aurora replicas
    ๐Ÿ’ฅ Impact GDPR Chapter V violation (international transfer without adequacy/safeguards); Schrems II implications
    ๐Ÿ“Š Likelihood Low (requires misconfiguration, but multi-region is complex)
    โš ๏ธ Risk Score 6.5/10 MEDIUM
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1537 Transfer Data to Cloud Account
    ๐Ÿ›ก๏ธ Planned Controls FUT-029: Geo-fenced replication (user PII stays in eu-west-1), AWS Config rules enforcing data residency, SCP preventing PII table replication to non-EU regions, automated compliance drift detection

    The platform's AI supply chain extends beyond npm packages to foundation models, training data, and MCP tool ecosystems โ€” each a potential vector for subtle, high-impact compromise.

    flowchart TD
    subgraph AI_SUPPLY_CHAIN["๐Ÿค– AI Supply Chain Attack Surface"]
    direction TB
    SC1["๐Ÿง  Foundation Model Updates<br/>Behavioral regression on upgrade"]
    SC2["๐Ÿ“ฆ MCP Server Dependencies<br/>Tool-level supply chain"]
    SC3["๐Ÿ“š Training Data Provenance<br/>Poisoned public datasets"]
    SC4["๐Ÿ”ง Prompt Template Integrity<br/>Workflow instruction tampering"]
    SC5["๐ŸŒ External API Dependencies<br/>Riksdag/SCB/IMF availability"]
    end

    subgraph GOVERNANCE_THREATS["โš–๏ธ AI Governance Threats"]
    direction TB
    AG1["๐Ÿ“œ EU AI Act Non-Compliance<br/>Regulatory classification change"]
    AG2["๐ŸŽฏ Model Behavior Drift<br/>Post-update output degradation"]
    AG3["๐Ÿ”„ Vendor Lock-In Exploitation<br/>AWS service discontinuation"]
    AG4["๐Ÿ“Š Transparency Debt<br/>Unexplainable model decisions"]
    end

    subgraph SUPPLY_CONTROLS["๐Ÿ›ก๏ธ Supply Chain Controls"]
    direction TB
    SCC1["๐Ÿ“Œ Model Version Pinning<br/>+ regression testing"]
    SCC2["๐Ÿ” SLSA Level 3<br/>Build provenance"]
    SCC3["โœ… Output Regression Suite<br/>Golden-set validation"]
    SCC4["๐Ÿ“‹ AI Model Cards<br/>Transparency documentation"]
    SCC5["๐Ÿ”„ Multi-Model Fallback<br/>Provider redundancy"]
    end

    SC1 --> AG2
    SC2 --> AG1
    SC3 --> AG2
    SC4 --> AG1
    SC5 --> AG3

    AG1 -.->|mitigated by| SCC4
    AG2 -.->|mitigated by| SCC1
    AG2 -.->|mitigated by| SCC3
    AG3 -.->|mitigated by| SCC5
    AG4 -.->|mitigated by| SCC4

    style AI_SUPPLY_CHAIN fill:#fff3e0,stroke:#e65100,color:#000
    style GOVERNANCE_THREATS fill:#fce4ec,stroke:#c62828,color:#000
    style SUPPLY_CONTROLS fill:#e8f5e9,stroke:#2e7d32,color:#000
    Attribute Detail
    ๐ŸŽญ Threat Agent Model provider (unintentional), adversary targeting model training
    โš”๏ธ Attack Vector A Claude or Bedrock model update introduces subtle behavioral changes: different political framing, altered fact-selection preferences, or degraded neutrality in Swedish-language outputs
    ๐ŸŽฏ Target All AI-generated content (14 news workflows, translation, analysis)
    ๐Ÿ’ฅ Impact Gradual quality/neutrality degradation across all outputs; potentially undetected for days if regression is subtle
    ๐Ÿ“Š Likelihood Medium (model updates are frequent; political-content testing is specialized)
    โš ๏ธ Risk Score 7.0/10 HIGH
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1195.003 Compromise Hardware Supply Chain (adapted: model supply chain)
    ๐Ÿ›ก๏ธ Planned Controls FUT-030: Model regression test suite (golden-set political content), automated neutrality scoring on model upgrade, staged rollout (canary โ†’ full), model version pinning with explicit upgrade gates
    Attribute Detail
    ๐ŸŽญ Threat Agent Regulatory environment change
    โš”๏ธ Attack Vector EU AI Act enforcement classifies the platform's election forecasting or political analysis as "high-risk AI" (Annex III, Category 8: administration of justice/democratic processes), triggering mandatory conformity assessment, transparency obligations, and human-oversight requirements
    ๐ŸŽฏ Target Platform operational model, AI governance framework, compliance posture
    ๐Ÿ’ฅ Impact Mandatory conformity assessment, potential operational restrictions during compliance period, significant documentation/audit requirements
    ๐Ÿ“Š Likelihood Medium (political-analysis AI is an emerging regulatory gray area)
    โš ๏ธ Risk Score 6.5/10 MEDIUM
    ๐Ÿ—‚๏ธ MITRE ATT&CK N/A (regulatory threat)
    ๐Ÿ›ก๏ธ Planned Controls FUT-031: Proactive EU AI Act alignment (maintain documentation as if high-risk), model cards per Bedrock model, human-oversight architecture already designed, transparency reports, regular legal-counsel review of classification guidance
    Attribute Detail
    ๐ŸŽญ Threat Agent Supply-chain attacker, compromised open-source maintainer
    โš”๏ธ Attack Vector Compromise an MCP server dependency (riksdag-regering, scb, world-bank, or upstream npm packages) to inject malicious tool responses into agentic workflows
    ๐ŸŽฏ Target 14 agentic news workflows consuming MCP tool responses as trusted inputs
    ๐Ÿ’ฅ Impact Poisoned data flows through multiple workflows, generating and publishing manipulated content at scale
    ๐Ÿ“Š Likelihood Low-Medium (MCP ecosystem is young, rapidly evolving, less audited than mature npm packages)
    โš ๏ธ Risk Score 7.5/10 HIGH
    ๐Ÿ—‚๏ธ MITRE ATT&CK T1195.001 Compromise Software Dependencies and Development Tools
    ๐Ÿ›ก๏ธ Planned Controls FUT-032: MCP server integrity verification (SHA-pinned versions, SBOM tracking), response schema validation, anomaly detection on MCP responses, sandboxed tool execution, SLSA Level 3 provenance for all build inputs

    As Riksdagsmonitor expands to Nordic and EU parliaments, it enters a contested information environment where state-level actors actively seek to undermine democratic institutions.

    flowchart TB
    subgraph GEO_CONTEXT["๐ŸŒ Geopolitical Context (2026โ€“2037)"]
    direction LR
    GC1["๐Ÿ‡ท๐Ÿ‡บ Hybrid Warfare<br/>Information operations<br/>targeting Nordic democracies"]
    GC2["๐Ÿ‡จ๐Ÿ‡ณ Influence Operations<br/>United Front Work targeting<br/>diaspora communities"]
    GC3["๐Ÿด Non-State FIMI<br/>Coordinated inauthentic<br/>behavior networks"]
    GC4["๐Ÿค– AI-Powered IO<br/>Synthetic media &<br/>automated propaganda"]
    end

    subgraph PLATFORM_EXPOSURE["๐Ÿ“ก Platform Exposure Points"]
    direction LR
    PE1["๐ŸŒ 14-Language Surface<br/>Each language = unique<br/>disinformation vector"]
    PE2["๐Ÿ”ฎ Forecast Outputs<br/>Election predictions as<br/>influence leverage"]
    PE3["๐Ÿค Federation Trust<br/>Nordic/EU data mesh<br/>as attack vector"]
    PE4["๐Ÿ“Š Credibility Capital<br/>Platform trust as<br/>laundering vehicle"]
    end

    subgraph GEO_DEFENSES["๐Ÿ›ก๏ธ Geopolitical Defenses"]
    direction LR
    GD1["๐Ÿ” FIMI Detection (C20)<br/>Early-warning indicators"]
    GD2["๐Ÿ“ Source Grading<br/>Confidence-floor enforcement"]
    GD3["๐ŸŒ Per-Language Review<br/>Native-speaker verification"]
    GD4["โš–๏ธ Advisory-Only Output<br/>No accusatory attribution"]
    end

    GC1 --> PE3
    GC1 --> PE1
    GC2 --> PE1
    GC3 --> PE4
    GC4 --> PE2

    PE1 -.->|defended by| GD3
    PE2 -.->|defended by| GD2
    PE3 -.->|defended by| GD1
    PE4 -.->|defended by| GD4

    style GEO_CONTEXT fill:#ffebee,stroke:#b71c1c,color:#000
    style PLATFORM_EXPOSURE fill:#fff3e0,stroke:#e65100,color:#000
    style GEO_DEFENSES fill:#e8f5e9,stroke:#1b5e20,color:#000

    The 14-language surface creates asymmetric verification challenges: content in languages without native-speaker review capacity (Arabic, Chinese, Japanese, Korean, Hebrew) presents higher manipulation risk.

    Language Tier Languages Verification Capacity Manipulation Risk Control
    ๐ŸŸข Tier 1 โ€” Native Review Swedish (sv), English (en) Full native review Low Direct editorial oversight
    ๐ŸŸก Tier 2 โ€” Accessible Review Norwegian (no), Danish (da), Finnish (fi), German (de), French (fr), Spanish (es), Dutch (nl) Accessible via Nordic/EU network Medium Back-translation + network review
    ๐Ÿ”ด Tier 3 โ€” Limited Review Arabic (ar), Hebrew (he), Japanese (ja), Korean (ko), Chinese (zh) Limited native review capacity High Enhanced back-translation, automated semantic-similarity scoring, community verification pipeline

    Control ID Control Name Threat Addressed STRIDE Coverage Implementation Target Priority
    FUT-023 Party-Symmetry CI Gate (automated neutrality audit) F13: Neutrality Erosion T, R Q2 2026 ๐Ÿ”ด Critical
    FUT-024 Longitudinal Sentiment-Balance Monitoring F13: Neutrality Erosion T Q3 2026 ๐Ÿ”ด Critical
    FUT-025 Election Cooling-Period Protocol F14: Election Manipulation T, D Q3 2026 ๐Ÿ”ด Critical
    FUT-026 Completeness Audit (Riksdag decision coverage) F15: Accountability Suppression R, I Q2 2026 ๐ŸŸก High
    FUT-027 Multi-Source Cross-Validation on Ingest F16: Information Laundering S, T Q2 2026 ๐Ÿ”ด Critical
    FUT-028 Privacy-by-Design Architecture (no opinion storage) F17: Political-Opinion Inference I 2027 Q3 ๐Ÿ”ด Critical
    FUT-029 Geo-Fenced Replication (EU PII residency) F18: Data Residency Violation I 2027 Q4 ๐ŸŸก High
    FUT-030 Model Regression Test Suite (golden-set) F19: Model Behavioral Regression T Q2 2026 ๐ŸŸก High
    FUT-031 Proactive EU AI Act Alignment F20: Regulatory Reclassification โ€” Q4 2026 ๐ŸŸก High
    FUT-032 MCP Server Integrity Verification (SHA-pinned) F21: MCP Ecosystem Compromise S, T Q2 2026 ๐Ÿ”ด Critical
    STRIDE Category Democratic/Privacy Primary Control Secondary Control Monitoring
    Spoofing Multi-source cross-validation (FUT-027) MCP integrity verification (FUT-032) Source-grade monitoring, ingest anomaly alerts
    Tampering Party-symmetry CI gate (FUT-023), model regression suite (FUT-030) Election cooling protocol (FUT-025) Longitudinal sentiment monitoring (FUT-024)
    Repudiation Completeness audit (FUT-026) Immutable calibration ledger Decision-coverage gap alerts
    Info Disclosure Privacy-by-design (FUT-028), geo-fenced replication (FUT-029) Data minimization, auto-purge Privacy-impact continuous assessment
    DoS Election cooling protocol (FUT-025) Rate limiting, human-escalation gates Election-window monitoring escalation
    Elevation EU AI Act alignment (FUT-031) Neutrality-as-governance Regulatory landscape scanning

    Threat Horizon Likelihood (1-5) Impact (1-5) Risk Score Treatment
    Gradual neutrality erosion via AI drift H2 4 5 20 CRITICAL MITIGATE (FUT-023, FUT-024)
    Election-period forecast manipulation H2/H3 3 5 15 CRITICAL MITIGATE (FUT-025)
    Democratic accountability suppression H2 2 5 10 CRITICAL MITIGATE (FUT-026)
    Information laundering via platform credibility H2 2 4 8 HIGH MITIGATE (FUT-027)
    Political-opinion inference from usage H3 3 4 12 HIGH MITIGATE (FUT-028)
    Cross-region data residency violation H3 1 4 4 MEDIUM MITIGATE (FUT-029)
    Foundation model behavioral regression H2 3 3 9 HIGH MITIGATE (FUT-030)
    EU AI Act regulatory reclassification H3 3 3 9 HIGH MITIGATE (FUT-031)
    MCP tool ecosystem compromise H2 2 4 8 HIGH MITIGATE (FUT-032)
    quadrantChart
    title Future Threat Risk Heat Map
    x-axis "Low Likelihood" --> "High Likelihood"
    y-axis "Low Impact" --> "High Impact"
    quadrant-1 "๐Ÿ”ด Critical โ€” Immediate Action"
    quadrant-2 "๐ŸŸก High โ€” Plan Mitigation"
    quadrant-3 "๐ŸŸข Low โ€” Monitor"
    quadrant-4 "๐ŸŸก Medium โ€” Assess Controls"
    "Neutrality Erosion (F13)": [0.75, 0.95]
    "Election Manipulation (F14)": [0.60, 0.90]
    "Vote Data Tampering (F1)": [0.55, 0.92]
    "Agent Excessive Agency (F8)": [0.40, 0.90]
    "IAM Escalation (F7)": [0.35, 0.92]
    "RAG Poisoning (F5)": [0.55, 0.80]
    "Forecast Manipulation (F9)": [0.55, 0.80]
    "Information Laundering (F16)": [0.40, 0.78]
    "Accountability Suppression (F15)": [0.35, 0.85]
    "Model Regression (F19)": [0.55, 0.60]
    "MCP Compromise (F21)": [0.40, 0.72]
    "API Abuse (F10)": [0.55, 0.45]
    "Data Residency (F18)": [0.20, 0.65]
    "EU AI Act (F20)": [0.55, 0.55]

    Phase Trigger Activities Output
    Pre-Implementation Feature design finalized STRIDE analysis, attack tree construction, control design Feature-specific threat addendum
    During Implementation Code review, PR merge Security testing, SAST/DAST scanning, dependency audit Security test results, remediation items
    Post-Deployment Feature goes live Penetration testing, monitoring activation, alert tuning Deployment security report
    Ongoing Quarterly review Threat landscape update, control effectiveness assessment Updated risk scores, new mitigations
    KPI Target Measurement Method
    New feature threat coverage 100% STRIDE per component Feature threat model completeness
    Time to detect data manipulation < 15 minutes Integrity check monitoring
    Cross-workflow anomaly detection rate > 95% Consistency check pass rate
    Translation integrity score > 98% accuracy Back-translation verification rate
    Pipeline data freshness SLA < 24 hours Cache timestamp monitoring
    WebSocket connection security 100% TLS 1.3 Connection protocol audit
    RAG / Knowledge-Base source provenance (H3) 100% allow-listed Bedrock KB ingestion audit
    Bedrock Agent action-scope conformance (H3) 100% within least-privilege policy Agent action-group / guardrail audit
    Cognito MFA enrolment for authenticated tier (H3) 100% of accounts Identity provider compliance report
    IAM least-privilege drift (H3) 0 over-privileged roles IAM Access Analyzer findings
    Multi-region replication integrity (H3) 100% checksum match Cross-region reconciliation audit

    Scores split by horizon. H2 threats can materialise 2026โ€“2027 while the platform is still static; H3 threats only become live once managed AWS services are provisioned (2027+).

    Threat Horizon Likelihood (1-5) Impact (1-5) Risk Score Treatment
    Real-time vote data manipulation H2 3 5 15 CRITICAL MITIGATE (FUT-001, FUT-009)
    CIA pipeline cache poisoning H2 2 4 8 HIGH MITIGATE (FUT-002, FUT-003)
    Multi-workflow AI orchestration attack H2 2 4 8 HIGH MITIGATE (FUT-004)
    Translation integrity attack H2 3 3 9 HIGH MITIGATE (FUT-005)
    Dashboard rendering DoS H2 3 2 6 MEDIUM MITIGATE (FUT-007)
    Politician profile defacement H2 2 3 6 MEDIUM MITIGATE (FUT-006)
    EU Parliament API compromise H2 1 3 3 LOW ACCEPT + MONITOR (FUT-008)
    Lambda/IAM privilege escalation & data exfiltration H3 2 5 10 CRITICAL MITIGATE (FUT-015, FUT-016, FUT-020)
    Bedrock Agent excessive agency H3 2 5 10 CRITICAL MITIGATE (FUT-017)
    Bedrock Knowledge-Base / RAG poisoning H3 3 4 12 HIGH MITIGATE (FUT-011, FUT-012)
    SageMaker election-forecast manipulation H3 3 4 12 HIGH MITIGATE (FUT-018)
    Cognito account takeover & IDOR H3 3 3 9 HIGH MITIGATE (FUT-013, FUT-014)
    AppSync/API Gateway public-API abuse H3 3 2 6 MEDIUM MITIGATE (FUT-019)
    Multi-region failover & replication tampering H3 1 4 4 MEDIUM MITIGATE (FUT-021)
    Nordic/EU federation cross-jurisdiction integrity H3 2 3 6 MEDIUM MITIGATE (FUT-022)


    ๐Ÿ“‹ Document Owner: James Pether Sรถrling, CEO & CISO
    ๐Ÿ“„ Version: 2.1
    ๐Ÿ“… Last Updated: 2026-06-02 (UTC)
    โœ… Approved by: James Pether Sรถrling, CEO
    ๐Ÿ”„ Review Cycle: Quarterly (Feb, May, Aug, Nov)
    โฐ Next Review: 2026-09-02
    ๐Ÿข Owner: Hack23 AB (Org.nr 5595347807)
    ๐Ÿ“ค Distribution: Public
    ๐Ÿท๏ธ Classification: Confidentiality: Public Integrity: High Availability: High

    Version Date Author Changes
    2.1 2026-06-02 James Pether Sรถrling Added Democratic Integrity & Accountability Threats (F13-F16); Privacy/GDPR threats (F17-F18); Supply Chain & AI Governance threats (F19-F21); Geopolitical & FIMI section; extended controls FUT-023โ€“FUT-032; risk heat map; language-specific threat vectors table; 5 new mermaid diagrams
    2.0 2026-05-31 James Pether Sรถrling Major expansion: Three-Horizon framework, Crown Jewel analysis, Attack Trees, Kill Chain mapping, OWASP LLM Top 10, Political-Intelligence capabilities
    1.0 2026-04-15 James Pether Sรถrling Initial future threat model with STRIDE and basic scenarios

    ๐ŸŽฏ Framework Alignment:
    ISO 27001 NIST CSF 2.0 CIS Controls OWASP EU AI Act


    Baseline: the already-implemented IMF STRIDE coverage (T-IMF-01..07) lives in THREAT_MODEL.md ยงIMF. The rows below (T-IMF-F-01..08) add future-state threats that emerge when the runtime migrates to Lambda + Aurora โ€” they extend the baseline rather than replace it.

    Authoritative hub: analysis/imf/README.md ยท analysis/imf/agentic-integration.md ยท analysis/imf/indicators-inventory.json ยท analysis/imf/data-dictionary.md ยท .github/aw/ECONOMIC_DATA_CONTRACT.md

    ID Element STRIDE Description Likelihood Impact Mitigation
    T-IMF-F-01 IMF cache (Aurora) Tampering Vintage substitution attack โ€” older WEO vintage swapped for newer label LOW HIGH SHA-256 payload pin + immutable supersedes-chain + CloudTrail audit
    T-IMF-F-02 IMF egress path DoS Workflow exhausts IMF rate limit (~30 req/min) โ†’ blocks legitimate articles MEDIUM MEDIUM Cache-first; โ‰ค30 req/min self-imposed; exponential back-off; metric alarm
    T-IMF-F-03 IMF payload Repudiation Article cites "IMF projects 2.1% growth" without vintage label โ†’ unauditable MEDIUM MEDIUM economicProvenance row required for every economic claim; cite_text mandatory
    T-IMF-F-04 IMF Datamapper schema Tampering Upstream schema change between WEO Apr/Oct cycles silently corrupts cache LOW HIGH Version-pinned client guard; CI integration test against IMF sandbox
    T-IMF-F-05 IMF data licence Repudiation Article reuses IMF figure without attribution (licence violation) LOW MEDIUM Article footer template auto-emits IMF citation block; lint enforces
    T-IMF-F-06 IMF cache fallback Information disclosure Stale vintage served to readers as current LOW MEDIUM Vintage-age badge (yellow >3mo, red >6mo); ECONOMIC_DATA_CONTRACT v2.1 banned phrases
    T-IMF-F-07 IMF + SCB cross-validation Tampering IMF SWE figure diverges >0.3pp from SCB national-accounts (silent error) LOW MEDIUM Quarterly cross-validation worker opens editorial-review issue
    T-IMF-F-08 IMF script supply chain Elevation tsx scripts/imf-fetch.ts execution path tampered upstream LOW HIGH Script in-repo; reviewed; no dynamic eval; harden-runner egress audit
    Tactic Technique IMF-specific application
    TA0006 Credential Access T1552 Unsecured credentials Datamapper transport is unauthenticated; SDMX 3.0 uses an Azure APIM subscription key (IMF_SDMX_SUBSCRIPTION_KEY) stored only as a GitHub Actions secret (never on disk, never logged); rotation playbook in analysis/imf/agentic-integration.md
    TA0007 Discovery T1083 File and directory discovery Cache directory permissions (read-only to article workers)
    TA0009 Collection T1530 Cloud storage object Aurora row-level access controls
    TA0040 Impact T1485 Data destruction Supersedes-chain prevents destructive overwrite

    Egress hosts (allow-list): www.imf.org (Datamapper REST ยท WEO/FM, unauthenticated), api.imf.org (SDMX 3.0 REST ยท IFS/BOP/DOTS/GFS/PCPS/ER/MFS_IR/MFS_PR, subscription-key authenticated via the Azure APIM Ocp-Apim-Subscription-Key header / IMF_SDMX_SUBSCRIPTION_KEY secret). Both HTTPS-only; payloads are public macro statistics with no PII.

    Canonical rule. Every economic claim in a Riksdagsmonitor article cites an IMF dataflow first; World Bank citations are reserved for governance, environment and social residue (the classes IMF does not publish). SCB is the Swedish-specific ground truth layer. See ECONOMIC_DATA_CONTRACT.md v2.1 for the banned-phrase list and vintage discipline (>6 mo โ†’ annotation).


    ๐ŸŒ Platforms ๐Ÿ“ฆ Open-Source Projects ๐Ÿ›ก๏ธ Governance & Standards
    ๐Ÿ—ณ๏ธ Riksdagsmonitor โ€” Swedish Parliament intelligence
    ๐Ÿ‡ช๐Ÿ‡บ EU Parliament Monitor โ€” European coverage
    ๐Ÿ•ต๏ธ Citizen Intelligence Agency โ€” political-data engine
    ๐ŸŒ Hack23 AB โ€” corporate site
    ๐Ÿ“ฐ Hack23 Blog โ€” engineering & policy
    ๐Ÿ’ผ Hack23 on LinkedIn
    ๐Ÿ—ณ๏ธ Hack23/riksdagsmonitor
    ๐Ÿ•ต๏ธ Hack23/cia
    ๐Ÿ‡ช๐Ÿ‡บ Hack23/euparliamentmonitor
    ๐Ÿ”Œ Hack23/european-parliament-mcp
    โœ… Hack23/cia-compliance-manager
    ๐Ÿฅ‹ Hack23/black-trigram
    ๐Ÿ  Hack23/homepage
    ๐Ÿ›ก๏ธ Hack23 ISMS-PUBLIC โ€” public ISMS
    ๐Ÿ”’ Information Security Policy
    ๐Ÿค– AI Policy
    ๐Ÿงช Secure Development Policy
    ๐ŸŽฏ Threat Modeling Policy
    โš ๏ธ Vulnerability Management
    ๐Ÿท๏ธ Classification Framework

    OpenSSF Best Practices OpenSSF Scorecard ISO 27001:2022 NIST CSF 2.0 CIS Controls v8.1 Apache 2.0

    ๐Ÿ—ณ๏ธ Empower citizens ยท ๐Ÿ” Strengthen democratic accountability ยท ๐Ÿ•ต๏ธ Illuminate the political process

    ยฉ 2008โ€“2026 Hack23 AB (Org.nr 559534-7807) ยท Maintainer: James Pether Sรถrling, CISSP CISM