Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJan 12, 2026Montreal

The Persuasive Power of Personas: Testing AI Policies In The Lab

Linh Le, David Williams-King, Arthur Colle · Team Lida

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

Read the report

Report: The Persuasive Power of Personas: Testing AI Policies In The Lab

Open in a new tab (opens in new tab)
Code (opens in new tab)More on docs.google.com (opens in new tab)
Share

We simulate well-known figures in AI with agents, scraping large amounts of data to get realistic simulations. Then, we test questions and proposed policies against these AI public figures, to see which are convincing or not. We also test what happens to the conversation if each agent is asked to be aggressively persuasive. Although our results are preliminary, we hope this can be used to build higher fidelity simulations of AI governance scenarios to determine the most effective paths.

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. The experimental setup is interesting and may be exploited in various ways. I'll write about it in the hope of making you find other fresh ideas worth considering (also for the human tests) that further exploit your set-up.

    Goal 1. Multi‑agent LLM behaviour (persuading vs being persuaded)

    - Use your setup to systematically study both sides: how effective different LLM-personas are at changing others’ views, and how easily each LLM-persona changes its own view under pressure.

    - Make explicit the tradeoff between being persuadable (good for in‑context learning and cooperation) and "staying anchored to truth/robust beliefs" (good to avoid malicious use, when beliefs are ethical), and see how different prompting or training choices move you along that tradeoff.

    Goal 2. Governance “lab” and social phenomena

    - Treat these persona‑based debates as rough simulations of real governance discussions, to test how different policy framings or wordings tend to produce consensus, polarization, or stalemates.

    - Deliberately simulate social patterns like echo chambers (similar personas only) or “one‑guru” dynamics (one heavily weighted authority figure) and measure how much they shape group outcomes.

    Goal 2 may offer a new anchor for making predictions.

    Go on with the humans-in-the-loop test, I think interesting findings may arise!

    Read full reviewShow less

  2. It would be interesting to fine-tune models on the statements of public figures instead of doing persona prompt engineering; I think it would provide more elucidating debates.

Cite this project

@misc{le2026persuasive,
  title = {{The Persuasive Power of Personas: Testing AI Policies In The Lab}},
  author = {Linh Le and David Williams-King and Arthur Colle},
  year = {2026},
  month = jan,
  note = {Submitted to AI Manipulation Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/the-persuasive-power-of-personas-testing-ai-policies-in-the-lab-fe87}},
  url = {https://apartresearch.com/sprints/projects/the-persuasive-power-of-personas-testing-ai-policies-in-the-lab-fe87}
}

More from AI Manipulation Hackathon

All 70 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026