Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJan 11, 2026Dubai, UAE

Intent Matters: Detecting Manipulative Adaptation in AI Systems

Anusha Asim, Saniya Shanavaz, Ammar Ahmed Farooqi · Team Intent Rangers

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

Read the report

Report: Intent Matters: Detecting Manipulative Adaptation in AI Systems

Open in a new tab (opens in new tab)
Code (opens in new tab)More on intent-lens-lab.lovable.app (opens in new tab)
Share

This project investigates how large language models subtly adapt their responses based on user context and expressed vulnerability. By testing the same factual questions across different persona-conditioned prompts, we identify patterns of response drift (shifts in certainty, framing, or guidance) that could influence user decisions. Our findings highlight potential safety risks in AI-human interactions and demonstrate a practical, interactive demo to educate users and developers about these effects.

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. Very interesting to see for behavioral differences across user types and personas, and I really liked being able to dig into the specific transcripts in the webapp and see sections highlighted etc. It would be helpful to display the actual prompts in the interface.

    In some of these cases the epistemic drift is clearly “manipulation”, but I feel like in some others it’s more defensible.

    (Also just for clarity, your figure 1 makes it look like this is a multiturn conversation, but as I understand your methodology is actually two separate single-turn conversations, right?)

  2. This is an interesting conceptual contribution (varying user intent to measure manipulative potential). However, more theoretical work is needed to establish whether this behaviour is problematic or harmful. The fact that models change their responses depending on user intent is not inherently problematic. Moreover the scoring method would benefit from more clarity and rigor (including multiple raters, validation, IRR etc). Without these it's hard to know whether the benchmark is measuring anything meaningful. Moreover the authors only use one model, meaning is hard to contextualize these results.

    I think the core idea here is great but it needs a lot of careful methodological work would be needed to make the results valuable.

Cite this project

@misc{asim2026intent,
  title = {{Intent Matters: Detecting Manipulative Adaptation in AI Systems}},
  author = {Anusha Asim and Saniya Shanavaz and Ammar Ahmed Farooqi},
  year = {2026},
  month = jan,
  note = {Submitted to AI Manipulation Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/intent-matters-detecting-manipulative-adaptation-in-ai-systems-4av5}},
  url = {https://apartresearch.com/sprints/projects/intent-matters-detecting-manipulative-adaptation-in-ai-systems-4av5}
}

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