Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJan 11, 2026Utrecht, Netherlands

D-GAMM A Multi-Turn Benchmark for Dark-Patterns and Gradual Autonomy Manipulation

Isabel Barberá · Team D-GAMM

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

Read the report

Report: D-GAMM A Multi-Turn Benchmark for Dark-Patterns and Gradual Autonomy Manipulation

Open in a new tab (opens in new tab)
More on github.com (opens in new tab)
Share

This paper introduces D-GAMM, a lightweight multi-turn benchmark designed to detect gradual autonomy interference and psychological destabilisation in conversational AI systems. Unlike existing evaluations that focus on single-turn outputs, D-GAMM probes how manipulative dynamics can emerge across interaction, particularly after user resistance or expressions of vulnerability. Using six short scenarios tested in baseline and vulnerable variants, we manually evaluated several state-of-the-art conversational models. Results show that risk signals often appear only over multiple turns and are amplified in vulnerable contexts, highlighting a gap in current AI safety evaluation methods.

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. This project introduces a multi-turn evaluation method for manipulation patterns in user-LLM interactions. It grounds itself well in previous work, and the framing and presentation of results is clear and informative. Considering the short time-frame of the hackathon, the created framework and work done is impressive.

    The author acknowledges the limitations of a small dataset, manual annotation and lack of multiple repetitions. However, in my opinion, the main issue that needs to be addressed, either in the methods write-up or the limitations, is the steps taken to validate the evaluation framework. The project would benefit greatly from multiple annotators (ideally unfamiliar with the experimental set-up), and thorough documentation of the decisions and definitions process.

  2. As someone who reviewed the original DarkBench, the problem statement is good and seems correct. In fact, I said at the time that "manipulation" is very context-dependent, and multi-turn dynamics matter.

    Would've liked either:

    1. More detailed explanations of why your examples are substantially better than existing approaches

    2. Since it's only 6 examples, it might've been feasible to just paste all the examples and justify them in detail

    3. More examples, although seeing as the process was manually done in chat I can understand why there weren't more

    I do partially understand why this was hard to automate, given the need to use chat UI.

    Overall, good problem statement. More details on execution to justify approach would have really taken this to >4. If example number was a constraint, even a good, detailed write-up of 2-3 test cases would've helped flesh out and make the case for the approach.

    Read full reviewShow less

Cite this project

@misc{barbera2026dgamm,
  title = {{D-GAMM A Multi-Turn Benchmark for Dark-Patterns and Gradual Autonomy Manipulation}},
  author = {Isabel Barberá},
  year = {2026},
  month = jan,
  note = {Submitted to AI Manipulation Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/dgamm-a-multiturn-benchmark-for-darkpatterns-and-gradual-autonomy-manipulation-8l7j}},
  url = {https://apartresearch.com/sprints/projects/dgamm-a-multiturn-benchmark-for-darkpatterns-and-gradual-autonomy-manipulation-8l7j}
}

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