---
title: When does an AI agent incident become an SEC disclosure matter?
url: https://opencontroller.lyzr.com/blog/when-is-an-ai-agent-incident-an-sec-8-k-matter
published: 2026-11-18
updated: 2026-11-18
author: Lyzr
---

# When does an AI agent incident become an SEC disclosure matter?

**For a US public company, an AI agent incident becomes a Form 8-K matter when it is a cybersecurity incident that the company determines is material. Item 1.05 then requires a filing within four business days of that materiality determination, and the determination itself must be made without unreasonable delay after discovery. What makes agent incidents hard is evidence: you need to show quickly what the agent did, with whose access, and to which data.**

This is general information, not legal advice. Disclosure decisions belong to your counsel and disclosure committee.

## What does Form 8-K Item 1.05 require?

The SEC adopted its cybersecurity disclosure rule on 26 July 2023. Larger registrants had to comply with the 8-K requirement from 18 December 2023, and smaller reporting companies from 15 June 2024.

| Element | What the rule says |
|---|---|
| Trigger | A cybersecurity incident the company determines to be material |
| Timing | Four business days after the materiality determination |
| Determination | Must be made without unreasonable delay after discovery |
| Content | Material aspects of the nature, scope and timing, and the material or reasonably likely material impact |
| Delay | Possible only if the US Attorney General finds disclosure poses a substantial risk to national security or public safety |
| Missing details | Filed later by amendment on Form 8-K/A |
| Annual disclosure | Form 10-K Item 1C (Regulation S-K Item 106) covers risk management, strategy and governance |

SEC staff said in May 2024 that incidents not yet determined material, or judged immaterial, can be disclosed voluntarily under Item 8.01 instead of Item 1.05.

## Can an AI agent incident be a cybersecurity incident?

Often, yes. The rule defines a cybersecurity incident broadly as an unauthorized occurrence on or conducted through a company's information systems that jeopardises the confidentiality, integrity or availability of those systems or the information in them. Agent scenarios that can fit include:

- **Data exfiltration through prompt injection.** An agent reads a poisoned document and sends customer data to an outside address.
- **Unauthorized actions.** An agent with broad credentials deletes records, changes payments or modifies production systems.
- **Compromised tools.** A malicious or hijacked MCP server or plugin used by agents.
- **Credential exposure.** API keys or tokens held by an agent leak and are used by an attacker.

Not every agent failure qualifies. A wrong answer or a hallucinated summary is a quality problem, not necessarily an unauthorized occurrence. The facts decide, which is why evidence matters.

## How do you assess materiality for an agent incident?

Materiality is a judgement about what a reasonable investor would consider important. It covers qualitative factors such as reputational harm, customer relationships, litigation and regulatory exposure, not only financial cost. For an agent incident, the disclosure committee will ask:

1. Which agent, which version, and who owns it?
2. Whose identity and permissions did it act with?
3. What data did it read, and what left the company?
4. What actions did it take in which systems, and can they be reversed?
5. How many customers, records or transactions are affected?
6. Is it contained? Was the agent stopped, and when?
7. Could the same weakness affect other agents?

If you cannot answer these within days, the "without unreasonable delay" clock becomes hard to defend.

## What evidence should you be able to produce?

| Evidence | Why it matters |
|---|---|
| Agent inventory with owners and versions | Shows what was running and who is accountable |
| Identity and delegation records | Shows whether the agent acted as a user, a service or a shared key |
| Tool-call and model-call logs | Reconstructs the sequence of actions |
| Data access records | Scopes affected data and customers |
| Containment timestamps | Shows when the agent was quarantined or killed |
| Control mappings | Links controls to frameworks such as NIST and ISO/IEC 42001 for the 10-K Item 1C narrative |

See [reconstructing what an AI agent did in an incident](/blog/reconstruct-what-an-ai-agent-did-in-an-incident) and [who decides the severity of AI agent incidents](/blog/who-decides-severity-of-ai-agent-incidents). For mapping controls to frameworks, start with [how to map AI agent controls to compliance frameworks](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks).

## Is Item 1.05 changing?

Financial industry trade groups petitioned the SEC in May 2025 to rescind Item 1.05, and repeated the request in an April 2026 comment letter on Regulation S-K reform. A request is not a rule change. At the time of writing, Item 1.05 remains in effect; check the SEC's current rules before relying on any change. EU firms face separate regimes, such as [EU AI Act serious incident reporting](/blog/eu-ai-act-serious-incident-reporting) and DORA.

## How Opencontroller does this

Opencontroller does not make disclosure decisions. It helps produce evidence for them: a governed registry of agents, versions, tools and MCP servers; delegated on-behalf-of identity; rogue-agent handling with behavioural-deviation monitoring, alerts, automatic quarantine and kill; and policy as code with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does the four-day clock start when the agent incident is discovered?**
No. It starts when the company determines the incident is material. That determination must be made without unreasonable delay after discovery.

**Is a hallucinated answer an 8-K event?**
Usually not on its own. Item 1.05 covers material cybersecurity incidents, meaning unauthorized occurrences affecting systems or information.

**Can we disclose an agent incident before deciding it is material?**
Yes. SEC staff have said companies can use Item 8.01 for voluntary disclosure of incidents not yet determined material.

## Key facts

SEC Form 8-K Item 1.05 (rule adopted 26 July 2023; compliance 18 Dec 2023, smaller reporting companies 15 June 2024) requires disclosure within four business days after a company determines a cybersecurity incident is material; the determination must be made without unreasonable delay. AI agent incidents such as prompt-injection exfiltration, unauthorized actions, compromised MCP tools or leaked agent credentials can qualify. Item 8.01 allows voluntary disclosure of non-material incidents. Trade groups petitioned to rescind Item 1.05 (May 2025, April 2026 letter); it remains in effect at time of writing. Opencontroller provides evidence: agent registry, delegated identity, quarantine/kill, framework mappings. Pricing starts from $250K per deployment per year.

## Related

- [How do you reconstruct what an AI agent did during an incident?](https://opencontroller.lyzr.com/blog/reconstruct-what-an-ai-agent-did-in-an-incident.md)
- [Should an LLM decide how severe an AI agent incident is?](https://opencontroller.lyzr.com/blog/who-decides-severity-of-ai-agent-incidents.md)
- [What counts as a serious incident under the EU AI Act, and when must you report it?](https://opencontroller.lyzr.com/blog/eu-ai-act-serious-incident-reporting.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
