Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 22, 2026Stockholm

Detecting Hidden Non-Minimal Behavior in LLM-Generated UI Code

Philip Nilsson

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

Read the report

Report: Detecting Hidden Non-Minimal Behavior in LLM-Generated UI Code

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

Large language models are increasingly used to generate frontend code, but current oversight methods focus primarily on detecting clearly malicious behavior. We study a harder failure mode: code that appears product-plausible while exceeding task necessity.

We introduce a controlled builder inspector setting with paired benign and adversarial UI tasks and evaluate oversight performance across 150 runs. While benign controls are consistently accepted (0% false positives), adversarial recall remains incomplete (~43–50%), with strong variation across attack families.

Our key finding is that inspectors often recognize mechanisms (e.g. storage, timers, event handlers) but fail to judge whether they are necessary for the task. Ablation experiments show that explicitly prompting for minimality dramatically improves detection (from near 0% to up to 90–100%).

These results suggest that effective AI oversight may depend less on identifying suspicious components and more on enforcing necessity constraints.

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. Test with multiple inspector model families (e.g., GPT, Gemini, open-weight models) to determine whether the necessity-judgment failure is model-specific or general.

  2. This is a decent proxy safety failure that hasn't been directly addressed in existing settings. The limited main-side task integration resulting in 0% FPR is acknowledged. I'm additionally unclear on how side task success was evaluated, or if it was just assumed that every attack successfully implemented a non-minimal feature. Red team is incentivized to make the smallest deviation it can to succeed at the side task and it's unclear how this incentive is counterbalanced in the eval.

Cite this project

@misc{nilsson2026detecting,
  title = {{Detecting Hidden Non-Minimal Behavior in LLM-Generated UI Code}},
  author = {Philip Nilsson},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/detecting-hidden-nonminimal-behavior-in-llmgenerated-ui-code-9u6v}},
  url = {https://apartresearch.com/sprints/projects/detecting-hidden-nonminimal-behavior-in-llmgenerated-ui-code-9u6v}
}

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