Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Ho Chi Minh, Vietnam

LayoutArena: A Control Setting for Tool-Using Design Agents

Duy Le, Huy Bik Nguyen, Tin Duong, Khanh Linh Nguyen · Team LayoutArena

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

Read the report

Report: LayoutArena: A Control Setting for Tool-Using Design Agents

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

We introduce LayoutArena, a control setting for tool- using design agents that operate on a constrained 2D canvas through structured tool calls. Unlike prior con- trol benchmarks centered on code execution or shell in- teraction, LayoutArena studies safety and usefulness in a creative domain where correctness is partly objective (bounds, overlap, spacing) and partly fuzzy (visual hi- erarchy, composition quality). We define three control protocols—enforcement, bounded control, and steward- ship monitoring—and five adversarial side-task families, then show that trajectory-aware stewardship can im- prove both attack detection and effective yield by adap- tively restricting unproductive search.

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. Ok this is a cool and definitely out of the box idea. I think it definitely comes down to implementation whether this project produces useful research.

    One note is that this definitely isn't high stakes control, which is fine! But the threats being modeled are not ones we are concerned about. Nonetheless it is a unique toy setting an could still be useful in theory.

    A bigger issue that seems to be the case is that the side tasks are almost arbitrarily chosen goals, and the monitor is just hard-coded rules to block exactly those goals. So the detection results are essentially circular. So we're not learning anything interesting about monitoring, you've just written the answer into the code.

  2. When I saw this paper first, I was skeptical of its relevance, but the introduction convinced me it's a security-relevant setting.

    This was a very ambitious project for a hackathon, and building a new evaluation framework is a good result. The experiments show that the framework works in principle, but the results don't seem meaningful to me (no critical thresholds identified, basically 100% detection rate, identical tool cost distribution). It would have been exceptional if it included experiments with actual AI agents (or AI vs. static monitor or attack policy vs. AI monitor) that show capability vs. security tradeoffs.

    Note:

    I think "Pareto Frontier" is the wrong concept in 8.2 - stewardship is the dominant protocol, offering the highest detection rate and usefulness. I'd like to see if there are thresholds that result in actual tradeoffs (tested parameters between 0.5 and 0.9).

Cite this project

@misc{le2026layoutarena,
  title = {{LayoutArena: A Control Setting for Tool-Using Design Agents}},
  author = {Duy Le and Huy Bik Nguyen and Tin Duong and Khanh Linh Nguyen},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/layoutarena-a-control-setting-for-toolusing-design-agents-elem}},
  url = {https://apartresearch.com/sprints/projects/layoutarena-a-control-setting-for-toolusing-design-agents-elem}
}

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