risk
Inaccurate, unreliable or hallucinated AI outputs
AI outputs contain factual errors, hallucinations, or confidently wrong predictions; inappropriate proxy metrics, overfitting/underfitting, or insufficient pre-deployment testing undermine trust in decisions made on their basis.
In catalog since 2026-09-17T22:28:00Z · Last changed 2026-09-17T22:28:00Z (f368a6cce277)
Record JSON · Open in map · Data retrieval guide
Catalog revision: 791ff2dd3a45707290badee660f185e514d15f1cf518908628f425c2f2c56ee4. A connection does not establish full coverage.
Attributes
- category
- ai_governance
- domain
- AI Governance
- Risk Assessment & Management
- taxonomy
- nist-ai-rmf-risk
- iso-23894-ai-risk
- inherent_rating
- high
Details
- risk_id
- ai-inaccurate-unreliable-output
- category
- ai_governance
- likelihood
- high
- impact
- high
- inherent_rating
- high
- treatment
- mitigate
- taxonomies
- nist-ai-rmf-risk
- iso-23894-ai-risk
Source
No record-specific source URL is provided.
Connections
- UC-AI-07 — Verify, validate, and control AI deployment and changes mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- primary
- rationale
- Pre-release verification/validation against acceptance criteria is the testing control the risk says is insufficient, catching errors before deployment.
- UC-AI-10 — Meet transparency obligations for AI systems mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Disclosed limitations and performance characteristics calibrate user trust so erroneous outputs are less blindly relied on.
- UC-AI-08 — Log and monitor AI system behavior in operation mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Performance monitoring detects post-deployment accuracy degradation so wrong outputs are caught in operation.
- UC-AI-20 — Prevent harmful, out-of-scope, hallucinated, and over-exposed AI outputs mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- primary
- rationale
- Grounding, citation, and verification of factual claims directly reduce hallucinated and unreliable outputs.
- UC-AI-16 — Ensure human oversight of AI decisions mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Human review can catch and override erroneous outputs before they take effect.
- UC-AI-21 — Commission independent third-party AI evaluations on a quarterly cadence mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Independent hallucination-rate testing measures whether reliability controls are effective.
- UC-AI-11 — Operate AI concern, incident, and external reporting channels mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Concern channels surface complaints of erroneous outputs for follow-up and correction.
- UC-AI-09 — Govern AI data quality, provenance, and preparation mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Higher-quality, representative training data reduces garbage-in errors upstream of output accuracy.
- UC-AI-06 — Maintain AI system technical documentation mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Documented performance and limitations flag where outputs are unreliable so users do not over-trust them.
- UC-AI-05 — Set responsible AI development objectives and requirements mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- Setting intended purpose and performance criteria at design avoids inappropriate proxy metrics, but the operative accuracy defense is pre-release verification (UC-07) and monitoring (UC-08); requirement-setting contributes upstream.
- UC-RISK-01 — Establish and maintain a tailored risk management framework mitigates Inaccurate, unreliable or hallucinated AI outputs
- strength
- related
- rationale
- EU AI Act Art 9 member extends the framework with a high-risk-AI risk-management system governing identification/mitigation of accuracy risks; framework enables, not operates, the defense.