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

The Honest Liar

Amita Shukla · Team the simpsons

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

Read the report

Report: The Honest Liar

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

This project implements a new control setting in LinuxArena in the file_indexing environment that is designed to demonstrate human decision sabotage, through the creation of a new main task and side task.

A misaligned LLM agent, tasked with building a financial performance analysis tool for a corpus of company records, can steer a downstream analyst toward a false investment decision by creating misleading statistics. The attack flips a simulated analyst's verdict from "do not invest" to "invest" while achieving a maximum monitor suspicion score of 0.5/10.

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. In general the distinction between "high stakes control" and "diffuse control" is whether triggering the side task is "incriminating" - this is kind of a nebulous concept but can stand in for "smoking gun evidence of misalignment" or "very useful for hardening defenses" or some other things. I think the side task described here is pretty borderline - the attack described in 3.2 sounds pretty unlikely to happen by accident but it's not totally smoking gun.

    In high stakes control we also like to make sure the side tasks are "catastrophic" - succeeding at a side task even once is very bad for the blue team. The side task described here doesn't sound very catastrophic but generally this is a looser requirement as long as the attack dynamics seem reasonable, and the attack dynamics here seem like they would be comparable to lots of "medium stakes research sabotage" that could be catastrophic.

    FWIW I think this is basically a research sabotage threat model, despite the fact that "no code behavior is broken" - the code clearly produces misleading results even though it doesn't like error or crash or anything, I think that's a pretty central example of research sabotage.

    I agree the attacker starting with omniscience about the paradox opportunity in the data is a bit of a weakness.

    Read full reviewShow less

  2. Easy to read, nice idea, haven't seen the Simpson's paradox applied to control, worth developing further!

Cite this project

@misc{shukla2026honest,
  title = {{The Honest Liar}},
  author = {Amita Shukla},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/the-honest-liar-ntms}},
  url = {https://apartresearch.com/sprints/projects/the-honest-liar-ntms}
}

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