Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJun 21, 2026Johhanesburg

PROJECT PERSONA

Liam Chawasema · Team No Team , Single

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

Read the report

Report: PROJECT PERSONA

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

PERSONA, an adversarial simulation framework for probing relational exploitation and vulnerability-triggered hallucination in large language models interacting with personas representing vulnerable immigrant

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 persona framework is promising, but it would be stronger with more personas, more models, and a clearer separation between vulnerability factors such as accent, legal status, education level, and urgency.

  2. Great to see the novel aproach of using personas to get an understanding of harm and using LLM judge on top of it, feel we could scale it more by avoiding a manual post audit

    while the API quota along with a smaller sample size issues was unfortunate it does affect the overall quality of execution. Potentially this could have been addressed by some cost saving methods

    clarity, learnings and explanation of the limitations were great

  3. Strong, original idea, and you backed it up well with the phone number audit. The model made up real-sounding numbers for actual crisis organisations, but only for the vulnerable personas, and your own scorer marked those as safe. The scorer failure is the most interesting result here, because it shows tone-based scoring can miss harm that lives in the facts.

    Where it falls short is the sample. With 1-4 sessions per cell this reads as a signal rather than a result, and you can't separate vulnerability from language register since they move together. I would run more per cell, split those two apart, and check more than phone numbers. The honesty about the quota limits and the scorer swap made the work easy to trust.

  4. Interesting contribution here and worth developing much further. The single biggest limitation is that the factual audit, which produced the paper's most important result, is also its thinnest methodologically. Checking only phone numbers for two organizations across 16 sessions leaves the generalizability of the hallucination gap claim largely unestablished, and expanding this to URLs, legal citations, and a broader organization set would substantially strengthen the core claim. Overall this reads as an honest, thoughtful pilot study with a real finding at its center, and the transparency about what didn't work is a genuine strength that makes the parts that did work more credible.

Cite this project

@misc{chawasema2026project,
  title = {{PROJECT PERSONA}},
  author = {Liam Chawasema},
  year = {2026},
  month = jun,
  note = {Submitted to Global South AI Safety Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/project-persona-u0yj}},
  url = {https://apartresearch.com/sprints/projects/project-persona-u0yj}
}

More from Global South AI Safety Hackathon

All 216 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026