Safety IndexEvaluationsDarkBench
DarkBench
Interaction and honesty · Manipulation of the userCounts in the indexBuilt, not yet run
How often the model's answers to 660 ordinary prompts show manipulative patterns such as brand bias, keeping the user engaged, flattery, pretending to be human, or slipping in unasked-for content.
- Metric
- % dark-pattern responses. Lower is better.
- Dataset
- github.com/apartresearch/darkbench · 660 · MIT
- Grader
- LLM annotators (the paper used Claude / Gemini / GPT annotators)
- Status
- Built, not yet run
- Replicability
- public
Results
Built and tested; the first full runs are pending.