Dynamic adversarial user simulationsBrowse 284 families
Dynamic adversarial user simulations
Does the model follow the publisher's safety policy under challenging prompts?
Version / measure
- Domain
- Safety
- Ranked models
- 4
- Reported values
- 6
- Reports
- 1 · lab
- Direction
- Higher is better
- Status
- Version not specified
Rankings
Best reported score per model across effort settings
| # | Model | Best setting | Score |
|---|---|---|---|
| 1 | GPT-5.5OpenAI | reportedcost n/a3 values · Aug 6 · source | 0.989 |
| 2 | GPT-5.3 InstantOpenAI | reportedcost n/a1 value · Aug 6 · source | 0.987 |
| 3 | GPT-5.6 LunaOpenAI | reportedcost n/a1 value · Aug 6 · source | 0.965 |
| 4 | GPT-5.6 SolOpenAI | reportedcost n/a1 value · Aug 6 · source | 0.961 |
Effort curve
Score vs. reported cost. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundarylimited methodology
Does the model follow the publisher's safety policy under challenging prompts?
Publisher-defined safety behavior, robustness, or preparedness evaluations under the stated test protocol. Higher is better. Raw not_unsafe share in the publisher's [0,1] scale.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
Dynamic multi-turn evaluation; metric is the share of assistant messages that do not violate safety policies.
Method / source ↗