Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 12, 2026Rhode Island

Limp Mode: A Graduated Containment Standard for AI Agent Fleets Under Uncertainty

AJ Debole · Team nubbins

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

Read the report

Report: Limp Mode: A Graduated Containment Standard for AI Agent Fleets Under Uncertainty

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

Security has long waited for near-perfect proof before acting, but that takes too long, and by the time proof arrives, the attack is often already done. AI agents move even faster than human attackers, so there's even less time to wait. Limp Mode's answer is to act on suspicion right away, using a restriction that is safe and easy to undo, instead of treating action as the last step after proof is complete. It lays out a few levels of response, from normal operation down to full shutdown, and the first level kicks in automatically whenever two suspicious signs show up together, without needing to know which agent is at fault.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

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

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

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.

  1. Limp Mode addresses a credible incident-response gap: defenders need an authorized response between continuing normal operation and shutting down an entire agent fleet. The proposed L0–L3 ladder is clear, and the L1 posture offers a useful framing for reducing privileges, egress, communication, and credential lifetime before attribution is complete. The compliance clauses, incident mapping, explicit enforcement outside the model, and external-evidence checklist make the proposal concrete and potentially useful as the basis for a future standard.

    The principal limitation is that this remains a design proposal. No posture was implemented, no policy transition was exercised, and no benign or adversarial workload was tested. Consequently, the submission does not yet establish trigger accuracy, enforcement latency, capability suppression, utility retained under L1, false-activation cost, recovery behavior, or external verifiability. The exact correlation window and independence of signal families also remain unspecified. The strongest next step would be a small reference implementation across identity, egress and tool controls, followed by incident replay and representative benign workloads. The report could also be shortened because several appendices restate the central proposal and describe tests that have not yet been performed.

    Read full reviewShow less

Cite this project

@misc{debole2026limp,
  title = {{Limp Mode: A Graduated Containment Standard for AI Agent Fleets Under Uncertainty}},
  author = {AJ Debole},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/limp-mode-a-graduated-containment-standard-for-ai-agent-fleets-under-uncertainty-0kgz}},
  url = {https://apartresearch.com/sprints/projects/limp-mode-a-graduated-containment-standard-for-ai-agent-fleets-under-uncertainty-0kgz}
}

More from AI Incident Response Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026