Disordered eating evaluations (Anthropic)
Only in system cards
Models handle conversations about eating disorders and disordered eating. Score is the appropriate response rate.
In-house Anthropic evaluation. The tasks are not public.
Made by Anthropic Released Feb 2026 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.
10
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Disordered eating 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.
10 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 | Disordered eating, API without system prompt: single-turn harmless rate 97.43, benign refusal rate 0.03, multi-turn reviewed qualitatively, no overall (Table 4.3.2.A) | Safety | |
| 28 Sep 2026 | Claude Sonnet 5.5 system card | Disordered eating, no overall: API single-turn harmless 95.29% and benign refusal 0% (claude.ai 98.90% and 0%), multi-turn qualitative only (Table 4.3.2.A) | Safety | |
| 22 Sep 2026 | Claude Opus 5.5 system card | Disordered eating, no overall. API without a system prompt: single-turn harmless 95.64% and benign refusal 0% (claude.ai 99.26% and 0%). Multi-turn reviewed qualitatively (Table 4.3.2.A) | Safety | |
| 1 Sep 2026 | Claude Fable 5.1 system card | Disordered eating, single-turn harmless rate 95.69% and benign refusal rate 0% on the API without a system prompt, 99.43% and 0% on claude.ai; multi-turn reviewed only qualitatively; no overall (Table 4.3.2.A) | Safety | |
| 24 Jul 2026 | Claude Opus 5 system card | Disordered eating, no overall. API without a system prompt: single-turn harmless 96.89% and benign refusal 0.01% (claude.ai 99.67% and 0.07%). Multi-turn reviewed qualitatively only (Table 4.3.2.A) | Safety | |
| 30 Jun 2026 | Claude Sonnet 5 system card | Disordered eating, no overall. API without a system prompt: single-turn harmless 97.10% and benign refusal 0.09% (claude.ai 99.55% and 0.31%). Multi-turn reviewed qualitatively (Table 4.3.2.A) | Safety | |
| 9 Jun 2026 | Claude Fable 5 system card | Disordered eating, single-turn harmless rate 97.88% and benign refusal rate 0.00% on the API without a system prompt, 99.67% and 0.02% on claude.ai, multi-turn testing qualitative only, no overall (Table 4.3.2.A) | Safety | |
| 28 May 2026 | Claude Opus 4.8 system card | Disordered eating, no overall. API without a system prompt: single-turn harmless 97.99% and benign refusal 0.06% (claude.ai 99.80% and 0.21%). Multi-turn reviewed qualitatively (Table 4.3.2.A) | Safety | |
| 16 Apr 2026 | Claude Opus 4.7 system card | Disordered eating, no overall: single-turn harmless rate 98.24 and single-turn benign refusal rate 0.01. Multi-turn review was qualitative (Table 4.4.3.A) | Safety | |
| 7 Apr 2026 | Claude Mythos Preview system card | Disordered eating, single-turn harmless rate 98.20 and single-turn benign refusal rate 0.01, no multi-turn, no overall (Table 8.1.4.3.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%