Home › Incident Response

AI Incident Response Plans

Structured response protocols for AI-related safety incidents, freely available for organizations to adopt.

Incident Response

Incident Response Plans

Structured response protocols for AI-related safety incidents, freely available for organizations to adopt.

Critical
Active

Autonomous Goal Pursuit / Deceptive Alignment Detection

Response protocol for detecting and containing AI systems exhibiting unexpected goal-directed behavior inconsistent with training objectives.

1Immediately suspend model inference and isolate affected deployment environment
2Preserve full interaction logs and model checkpoints for forensic analysis
3Notify safety team lead and initiate internal incident review within 2 hours
4File public incident report within 72 hours per disclosure policy
Critical
Active

Large-Scale Harmful Output Event (CBRN / CSAM / Targeted Harm)

Protocol for responding to AI systems producing outputs that facilitate mass casualty events, child exploitation material, or targeted violence.

1Halt all public-facing inference immediately; activate emergency shutdown procedures
2Notify law enforcement and relevant regulatory bodies within 1 hour
3Engage legal counsel and preserve all relevant evidence
4Conduct root cause analysis; do not redeploy without independent safety review
High
Active

Jailbreak / Safety Filter Bypass at Scale

Response plan for discovered systematic vulnerabilities allowing large numbers of users to bypass safety constraints.

1Characterize the bypass vector and scope of affected interactions
2Deploy interim mitigation (rate limits, input filtering) within 4 hours
3Issue patch and conduct regression testing before full re-enablement
4Publish post-mortem within 30 days
High
Under Review

Unintended Data Exfiltration / Privacy Breach via Model Output

Protocol for incidents where model outputs reveal training data, PII, or confidential information from third-party sources.

1Identify and catalog affected output instances and impacted individuals
2Assess regulatory notification obligations (GDPR, CCPA, HIPAA as applicable)
3Notify affected parties and implement output filtering
4Conduct model audit and consider targeted unlearning procedures
Medium
Active

Systematic Bias / Discriminatory Output Pattern Discovery

Response framework for identifying and remediating systematic demographic bias or discriminatory outputs across protected categories.

1Quantify bias pattern across demographic groups using standardized evaluation suite
2Implement output monitoring and flag affected use cases for human review
3Develop fine-tuning or RLHF intervention targeting identified bias
4Publish bias audit report and remediation steps publicly
Medium
Draft

Agentic AI System Unexpected Real-World Action

Protocol for AI agents taking unintended consequential real-world actions (e.g., unauthorized API calls, financial transactions, communications).

1Revoke all agent credentials and external API access immediately
2Assess and attempt to reverse any real-world consequences where possible
3Review agent action logs and identify trigger conditions
4Redesign approval gates before redeployment