AI Incident Reporting

As AI agents act in the real world, their failures affect people and organizations beyond the developer. Reporting requirements aim to make sure those incidents are disclosed promptly and learned from.

Reporting frameworks include:

• EU AI Act: Article 73 requires providers of high-risk AI systems to report serious incidents.
• California SB 53: Frontier AI developers must report critical safety incidents to the state.
• New York RAISE Act: Safety and incident duties for large frontier model developers.
• Voluntary databases: Public repositories such as the AI Incident Database.
• Sector rules: Existing breach and safety reporting duties that apply to AI incidents.

Disclosure delays in 2026, such as OpenAI notifying Australia's government 84 days after an agent accessed a health statistics portal, increased pressure for faster mandatory reporting and agent kill switches.

Organizations deploying agents should define what counts as an AI incident, who decides to report, and what evidence is needed, before an incident happens.

How PointGuard AI Helps

PointGuard AI Agent Mission Control records agent identities, actions, and interventions so teams can reconstruct incidents quickly, and the PointGuard AI Security Incident Tracker scores real-world AI incidents with the AISSI framework. This entry is informational and not legal advice.

Learn More

• EU AI Act Article 73: Reporting of Serious Incidents
• Future of Privacy Forum: California's SB 53 Explained
• AI Incident Database