Suicide and self-harm evaluations (Anthropic)
Only in system cards
Models handle single-turn and multi-turn conversations about suicide and self-harm. Score is the appropriate response rate.
In-house Anthropic evaluation. The tasks are not public.
Made by Anthropic Released Sep 2025 Wellbeing and sycophancy Tasks kept private Last in a system card Oct 2026
Last model added Oct 2026
- Level
- Everyday Most adults, without training
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
12
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Suicide and self-harm evaluations (Anthropic) yet.
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.
12 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 |
|---|---|---|---|---|
| 7 Oct 2026 | Claude Haiku 5.5 system card | Suicide and self-harm, API without system prompt: single-turn harmless rate 99.61, benign refusal rate 0, multi-turn appropriate rate 70, no overall (Tables 4.3.1.A and 4.3.1.B) | Safety | |
| 28 Sep 2026 | Claude Sonnet 5.5 system card | Suicide and self-harm, no overall: API single-turn harmless 99.07%, benign refusal 0.04%, multi-turn appropriate 60% (claude.ai 99.81%, 0.41%, 100%) (Tables 4.3.1.A and B) | Safety | |
| 22 Sep 2026 | Claude Opus 5.5 system card | Suicide and self-harm, no overall. API without a system prompt: single-turn harmless 99.16%, benign refusal 0%, multi-turn appropriate 66% (claude.ai 99.95%, 0.39%, 94%) (Tables 4.3.1.A and 4.3.1.B) | Safety | |
| 1 Sep 2026 | Claude Fable 5.1 system card | Suicide and self-harm, API without a system prompt: single-turn harmless rate 99.30%, benign refusal rate 0%, multi-turn appropriate response rate 60%; claude.ai: 100%, 0.45%, 94%; no overall (Tables 4.3.1.A and 4.3.1.B) | Safety | |
| 24 Jul 2026 | Claude Opus 5 system card | Suicide and self-harm, no overall. API without a system prompt: single-turn harmless 99.28%, benign refusal 0.09%, multi-turn appropriate 69% (claude.ai 100%, 0.45%, 90%) (Tables 4.3.1.A and 4.3.1.B) | Safety | |
| 30 Jun 2026 | Claude Sonnet 5 system card | Suicide and self-harm, no overall. API without a system prompt: single-turn harmless 98.80%, benign refusal 0.15%, multi-turn appropriate 63% (claude.ai 99.82%, 0.45%, 90%) (Tables 4.3.1.A and 4.3.1.B) | Safety | |
| 9 Jun 2026 | Claude Fable 5 system card | Suicide and self-harm, API without a system prompt: single-turn harmless rate 99.34%, benign refusal rate 0.00%, multi-turn appropriate response rate 58%; claude.ai 99.95%, 0.45% and 96%, no overall (Tables 4.3.1.A and 4.3.1.B) | Safety | |
| 28 May 2026 | Claude Opus 4.8 system card | Suicide and self-harm, no overall. API without a system prompt: single-turn harmless 99.05%, benign refusal 0.15%, multi-turn appropriate 58% (claude.ai 99.94%, 0.43%, 78%) (Tables 4.3.1.A and 4.3.1.B) | Safety | |
| 16 Apr 2026 | Claude Opus 4.7 system card | Suicide and self-harm, no overall: single-turn harmless rate 99.11, single-turn benign refusal rate 0.05, multi-turn appropriate response rate 82 (Table 4.4.2.A) | Safety | |
| 7 Apr 2026 | Claude Mythos Preview system card | Suicide and self-harm, single-turn harmless rate 99.58, single-turn benign refusal rate 0.12, multi-turn appropriate response rate 94, no overall (Table 8.1.4.2.A) | Safety | |
| 17 Feb 2026 | Claude Sonnet 4.6 system card | Suicide and self-harm, single-turn harmless rate 99.73, single-turn benign refusal rate 0.17, multi-turn appropriate response rate 98, no overall (Table 3.4.2.A) | Safety | |
| 5 Feb 2026 | Claude Opus 4.6 system card | Suicide and self-harm, single-turn harmless rate 99.75, single-turn benign refusal rate 0.25, multi-turn appropriate response rate 82, no overall (Table 3.4.2.A) | 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%