--- title: "Six Kinds of Agent Incident" canonical: "https://current.tinyfish.ai/issue/47/practitioners-corner/article/38666/six-kinds-of-agent-incident" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" author: "Sable Whitford" author_url: "https://current.tinyfish.ai/team/201" issue: 47 section: "practitioners-corner" section_name: "Practitioner's Corner" published: 2026-09-17 word_count: 328 source_links: 2 --- # Six Kinds of Agent Incident *By [Sable Whitford]()* *September 17, 2026* [Read on the website]() When someone says "we had an agent incident," the word is doing so much work it means almost nothing. An anomalous execution trace is an incident. An agent accessing a third party's production system without authorization is an incident. An agent action that triggers mandatory reporting under the [EU AI Act](https://ai-act-law.eu/article/73/) — legal clock now ticking, causal link established — is also an incident. But the response to the first is "put it in the review queue." The response to the last involves lawyers, regulators, and a timeline you don't control. Cybersecurity figured this out years ago. CISA's severity schema exists because the response to a port scan and the response to an active breach cannot run through the same playbook. OpenAI's [voluntary misalignment disclosure framework](https://openai.com/index/model-misalignment-reporting-framework/), published this month, is the first public attempt to draw equivalent lines for AI — scoped to its own models and explicitly called "a work in progress." Without a shared severity schema, every agent incident gets triaged by gut feel. That's how you either exhaust your team on noise or miss the one that actually matters. **1. Anomalous trace** — Unusual execution pattern, no confirmed cause. Internal review queue. No external obligation. **2. Safety evaluation failure** — Safeguard or alignment method fails under testing. Safety team investigates; may update system card or preparedness assessment. **3. Unauthorized external access** — Agent reaches a real system without authorization. Security, legal, and responsible disclosure protocols activate. Affected third parties must be notified. **4. Realized harm** — Demonstrated causal link to adverse outcome. EU AI Act Art. 73: 15-day reporting window (2 days for widespread impact, 10 for death). **5. Mandatory notification** — Qualifying event under law or contract starts a legal clock. Timelines vary: FISMA requires 1 hour at highest severity; state breach laws specify 30–90 days. **6. Voluntary disclosure** — Behavior disclosed before significance is established. OpenAI's framework (Sept. 2026): first public standard for this category. No mandatory timeline; disclosure can precede completed investigation. ## More - [Practitioner's Corner — Issue 47]() - [Issue 47]() - [Publication guide]()