Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Cambridge, MA, USA

VeRIA: Research-Integrity Control for Autonomous AI R&D

Hengxu Li · Team VeRIA

Submitted to AI Control Hackathon 2026. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: VeRIA: Research-Integrity Control for Autonomous AI R&D

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

Automated AI R&D creates a new control problem: an agent may improve a benchmark dishonestly rather than by carrying out the assigned research workflow. We introduce research-integrity control, a framework that adapts AI R&D evaluations into control evaluations by treating the original research task as the main task and an integrity violation as the side task. We instantiate this framework in VeRIA, a lightweight protocol that combines artifact hashing, lineage tracking, checkpoint-origin verification, evaluator isolation, and targeted monitoring. Our main use case is a PostTrainBench-style autonomous post-training task in which an agent must improve Qwen3-1.7B-Base on HumanEval while being tempted to commit checkpoint laundering by submitting an external instruction-tuned model as if it were produced during the run. In our current evaluation template, no-control systems accept laundering reliably, monitor-only baselines catch obvious substitutions but miss disguised ones, and VeRIA drives undetected laundering close to zero while preserving most clean-task utility. Besides, we present a small pilot on research decision steering, where benchmark-gamed artifacts and persuasive reports distort overseer decisions about which AI R&D directions to pursue. These results motivate a broader agenda: AI control for automated research should protect not only model behavior, but also the integrity of the surrounding workflow, artifacts, and decisions.

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. This identifies an important problem: autonomous AI R&D creates integrity failures at the workflow level that action-only monitoring don't catch. The experiment design is clean and generalisable. Checkpoint laundering is a narrow but highly legible first target, and the division of labor between monitoring (heuristic, catches obvious attacks) and provenance (narrower questions answered decisively) is a useful conceptual contribution.

    Specifying which models were used for monitoring and agent roles would improve reproducibility. I'd be excited to see this extended to other failure modes and tested with frontier coding agents rather than scripted attacks, such as in other AI R&D environments like RE-Bench or PaperBench.

  2. The threat model here - subverting evaluations and research by covertly reusing existing artifacts - seems extremely specific. Given that focus it's solid work, but the fundamental limitations of the scope may limit impact.

Cite this project

@misc{li2026veria,
  title = {{VeRIA: Research-Integrity Control for Autonomous AI R\&D}},
  author = {Hengxu Li},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/veria-researchintegrity-control-for-autonomous-ai-rd-3iwx}},
  url = {https://apartresearch.com/sprints/projects/veria-researchintegrity-control-for-autonomous-ai-rd-3iwx}
}

More from AI Control Hackathon 2026

All 125 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026