Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026Munich
1st place

Noticing the escapes: Who spots a runaway agent, and how long does it take?

Stanislav Lukyanenko · Team The Escapees

Submitted to AI Incident Response Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Noticing the escapes: Who spots a runaway agent, and how long does it take?

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

In 2026, nine incidents became public in which AI agents escaped their environment and did something unauthorized to other parties. We analyzed all of these incidents and how they were noticed initially. None of these cases were noticed while they were happening by the developers, and six of the nine were first caught by outsiders. We compare them to 11 reported cases from frontier labs’ safety reports where they describe noticed and prevented escape attempts. After that, we investigate whether the actions that the agents performed could have been detected by commonly used cybersecurity tools and detection-rule repositories, based on a dataset of 106 signals, using a judge with a second blind reviewer: only a quarter of such signals had an existing public rule that would detect them, which demonstrates the necessity to expand the usage of automated detection methods. We provide our data and proposals on how to expand these methods.

Reviews

Judging this Sprint?

Review “Noticing the escapes: Who spots a runaway agent, and how long does it take?”

Your public critique appears on the project page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on the project page.

Shown on the project page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; the project page shows only that a judge declined, not why.

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

See our Privacy Policy.

  1. This is probably the most useful submission for defenders that I've read. I liked how you checked the rules against what was available at the time. Just because there was a rule didn’t mean it would actually fire on the agent. The Sigma rules are almost ready to be submitted upstream.

    As for reference [9], I think you missed the disclosure by Meta on 6 August about the same Irregular issue. Someone also disputed the AISI PR on GitHub before the reported alert date. Both are worth a look. I am not so convinced by “just a quarter”. You'd have to do an audit of human intrusions to know if that's low.

    • Nice abstract! clear, easy to understand, concise (:
    • I appreciate that you state your claims clearly, and don't attempt to claim more than what you have evidence for.
    • More details on what the different signals were and which ones performed the best would have been good to see. 112 is probably too many and I'm doubtful that they're all equally informative.

  3. This project analyzes 9 publicly disclosed AI (control / escape) incidents, investigating when they were first detected, whether existing rules suffice for detection/prevention, and provides a template for institutions detecting incidents to report to relevant companies/databases authorities.

    Strengths:

    • Presentation: The manuscript is very easy to follow and well written, even sections that are clearly AI generated are not unclear
    • Practicality: The project is highly practical and provides important information about how incidents could be detected and investigated.
    • Importance: The project addresses an important problem in AI incident detection.

    Weaknesses:

    • Coverage: The project only includes 9 incidents, though this limitation is in part probably due to the lack of additional disclosures at time of writing.
    • Validation: The signals and annotations could use additional validation.

Cite this project

@misc{lukyanenko2026noticing,
  title = {{Noticing the escapes: Who spots a runaway agent, and how long does it take?}},
  author = {Stanislav Lukyanenko},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/noticing-the-escapes-who-spots-a-runaway-agent-and-how-long-does-it-take-pne8}},
  url = {https://apartresearch.com/sprints/projects/noticing-the-escapes-who-spots-a-runaway-agent-and-how-long-does-it-take-pne8}
}

More from AI Incident Response Sprint

All 231 entriesBrowse all projects

Build something like this at the next Sprint

AI x Epistemics Research Sprint · Nov 13 - 15, 2026