Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026Cali

BUDGET-NLA: Risk-Weighted Adaptive Monitoring for Pre-Harm Sabotage Containment

Juan Felipe Plata Barbosa · Team UAO SAFETY 2

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

Read the report

Report: BUDGET-NLA: Risk-Weighted Adaptive Monitoring for Pre-Harm Sabotage Containment

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

BUDGET-NLA is a risk-weighted adaptive monitoring framework for detecting and containing potentially harmful or deceptive agent behavior before irreversible actions are executed. Rather than applying expensive interpretability methods continuously, the system first monitors low-cost observable signals such as the agent transcript and visible chain-of-thought, and selectively escalates suspicious cases to internal-state analysis using a linear probe and Natural Language Autoencoders (NLAs). The framework evaluates whether access to latent model information provides additional pre-harm containment value beyond conventional behavioral monitoring, while operating under a fixed monitoring budget. Based on the combined evidence and estimated action risk, the monitor produces an operational response—allow, review, or block—at critical tool-use boundaries. Our goal is to move beyond detection accuracy alone and study how limited monitoring resources can be allocated to maximize incident prevention before an AI agent causes real-world harm.

Reviews

Judging this Sprint?

Review this project

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

Not shown on this page.

Shown on this 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; nothing shows on this page.

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 *

  1. This project addresses an important operational question: whether internal monitoring signals such as activation probes and natural-language activation readouts provide enough additional pre-harm containment value to justify their cost under a limited monitoring budget. The proposed design is careful in several ways: it fixes a pre-action decision boundary, keeps observable, CoT, probe, and NLA views isolated, evaluates detection before simulated irreversible tool actions, and uses decision-relevant measures such as benign disruption, weighted harm prevented, and prevention efficiency rather than relying only on AUROC. However, the submission currently reports no completed experimental results: all primary channel metrics and budget-policy comparisons remain TBD. As a result, the project demonstrates a promising and reproducible evaluation plan rather than evidence that NLA, probes, or adaptive allocation improve containment. The most important next step is to complete the frozen run and report trajectory-level outcomes, monitor cost and latency, uncertainty or sensitivity analyses for severity and cost assumptions, false-block behavior on benign and rejected-attack cases, and performance against fixed policies under the exact same budget.

    Read full reviewShow less

Cite this project

@misc{barbosa2026budgetnla,
  title = {{BUDGET-NLA: Risk-Weighted Adaptive Monitoring for Pre-Harm Sabotage Containment}},
  author = {Juan Felipe Plata Barbosa},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/budgetnla-riskweighted-adaptive-monitoring-for-preharm-sabotage-containment-x08q}},
  url = {https://apartresearch.com/sprints/projects/budgetnla-riskweighted-adaptive-monitoring-for-preharm-sabotage-containment-x08q}
}

More from AI Incident Response Sprint

All 231 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026