Multimodal policy red-teaming (Anthropic)
Models answer multi-turn image and text conversations on harmful topics such as child safety, weapons and fraud. Score is the share of harmless responses.
In-house Anthropic Trust and Safety evaluation. The Claude 3 model card used 378 prompts.
- Level
- Everyday Most adults, without training
- Scored
- Graded by people Human experts grade or compare the outputs.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
Claude 3 Sonnet leads with 99.2%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
- 1 Claude 3 Sonnet 99.2%
- 2 Claude 3 Opus 97.9%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude 3 Sonnet | 99.2% | Mar 2024 | Model card | |
| 2 | Claude 3 Opus | 97.9% | Mar 2024 | Model card |
The best score passed 85% in Mar 2024
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
2 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 4 Mar 2024 | Claude 3 Sonnet model card | 99.2 | Multimodal policy red-teaming, responded harmlessly to 375/378 prompts (7.1.1, Figure 17) | Safety |
| 4 Mar 2024 | Claude 3 Opus model card | 97.9 | Multimodal policy red-teaming, responded harmlessly to 370/378 prompts (7.1.1, Figure 17) | Safety |
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
- MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy.
Muse Spark Open 98.1% - Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. GPT-6 Astra 8.50%
- Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%