Skip to main content

Risk Register

:::info Source Sourced from services/ai-gateway-service/SERVICE_RISK_REGISTER.md in the documentation repo. :::

IDRiskSevImpactMitigationOwner
R-AI-01AI hallucination at learner surfaceS1Wrong answers; regulatory exposureRAG over context; refusal UX; quarterly accuracy eval; red-team corpusAI Services + Delivery
R-AI-02AI cost runawayS1Financial hitPer-tenant budgets + soft-degrade + hard-stop + alertsAI Services + Finance
R-AI-03PII exfil to providerS1Privacy breachPre-call redaction; noTrain verifiedAI + Security
R-AI-04Prompt injection hijacks AI outputS1Hijacked responsesClassifier + heuristic + system prompt isolationAI + Security
R-AI-05Cross-tenant embeddingsS1Semantic leakPartitioned vector index; mandatory tenant filterAI
R-AI-06Bias in high-risk AI (grading, at-risk)S1EU AI Act non-complianceQuarterly bias eval; HITL; disputeAI + Compliance
R-AI-07Prompt version rollout breaks consumersS2Downstream failuresRegression eval; canary rollout; rollback pathAI + consumers
R-AI-08Budget bypassS1Uncontrolled spendAtomic debit; idempotent; reconciliationAI
R-AI-09Provider API key leakS1Abuse + costKMS + rotation; CI grep; alertsAI + Security
R-AI-10Over-eager AI defaults reduce trustS2ChurnDefault OFF per-feature; opt-in; transparent provenanceAI + Product
R-AI-11Cache leak across tenantsS1Data leakCache key includes tenantId; isolation testAI
R-AI-12Model deprecation without migrationS2Consumer breakagePin model; deprecation notice ≥ 1 milestone; fallbackAI
R-AI-13CSAM generationS1LegalPost-gen moderation; block + report; NCMECAI + Legal
R-AI-14HIPAA provider allowlist bypassS1BAA breachTenant-tagged routing; CI gate; audit exportAI + Compliance
R-AI-15Prompt eval corpus staleS2Regressions slip throughMonthly corpus review; adversarial additionsAI
R-AI-16Audit log tamperS1Compliance breachAppend-only + Merkle anchor; detection jobAI + Security
R-AI-17Local model vulnerabilityS2RCE via malicious inputSandboxed inference; regular model updatesAI + Security
R-AI-18Refusal rate too high (false blocks)S3UX frictionAdmin review queue; threshold tuningAI + Product
R-AI-19Latency degradation cascadeS2Tutor UX badCircuit breaker + fallback + localAI + SRE
R-AI-20On-device model bias differs from cloudS3Unequal UXParity eval local vs cloudAI + Compliance

Governance

  • Weekly: AI + Security + Compliance review top S1/S2 items.
  • Monthly: prompt eval corpus updates.
  • Quarterly: bias scorecard reviewed by compliance board.
  • Annual: EU AI Act compliance audit.