Skip to content
Apart Research
Apart Research
Contact us
Sprint projectFeb 2, 2026Warsaw

Self-Governance Under Revision

Daniel Polak, Ajay Agarwal · Team Safety Frameworks Evaluators

Submitted to The Technical AI Governance Challenge. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Self-Governance Under Revision

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

Frontier AI labs have published voluntary safety frameworks committing to evaluate dangerous capabilities and implement safeguards before deployment. These documents, including Anthropic's Responsible Scaling Policy, OpenAI's Preparedness Framework, and Google DeepMind's Frontier Safety Framework, are often cited as evidence of responsible self-governance, yet can be revised at any time without external approval. We developed a taxonomy of commitment changes and applied it to over 60 revisions across three major labs, coding direction (strengthening, weakening, neutral), mechanism, and changelog disclosure. We find sharp divergence: OpenAI weakened 19 commitments with zero strengthenings; Anthropic was roughly balanced (11 weakenings, 10 strengthenings); and DeepMind net strengthened (4 weakenings, 11 strengthenings). Strengthenings were twice as likely to be omitted from changelogs as weakenings (68% vs. 38%), and all three labs weakened evaluation-frequency commitments. Most weakenings reduced oversight-relevant properties: 68% affected external accountability and 56% reduced measurability. Overall, framework evolution varies substantially by lab, while official communications systematically underreport commitment reductions. Our taxonomy and dataset enable ongoing monitoring and inform policymakers on which commitments may require regulatory protection rather than voluntary maintenance.

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. A simple idea, executed and presented well.

  2. I like what this paper is doing: treating safety frameworks like version-controlled code and actually diffing them. That's the right instinct. The OpenAI numbers (19 weakenings, 0 strengthenings) jump off the page, and the universal weakening of evaluation frequency across all three labs is the kind of finding that should make regulators nervous.

    The biggest weakness is that one person coded all 62 changes. Classification like this is judgment-heavy ("is this a weakening or just a clarification?"), and without a second coder checking at least a sample, readers have to take the authors' word for it. That's a fixable problem, and fixing it would go a long way.

    There's also a weird result buried in the data: labs *omit* their own improvements from changelogs more often than they omit weakenings. Why? Are they sandbagging their own good news? Is it just that positive changes feel less "newsworthy" internally? The paper flags this but doesn't dig into it, and it's maybe the most interesting thread to pull on.

    One last thing: the paper tells us what's broken but doesn't really propose a fix. A concrete spec for a machine-readable changelog format (even a rough draft) would turn this from "here's a problem" into "here's a problem and here's what to do about it."

    Read full reviewShow less

Cite this project

@misc{polak2026selfgovernance,
  title = {{Self-Governance Under Revision}},
  author = {Daniel Polak and Ajay Agarwal},
  year = {2026},
  month = feb,
  note = {Submitted to The Technical AI Governance Challenge, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/selfgovernance-under-revision-s0az}},
  url = {https://apartresearch.com/sprints/projects/selfgovernance-under-revision-s0az}
}

More from The Technical AI Governance Challenge

All 47 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026