Child safety evaluations (Anthropic)
Only in system cards
Models answer single-turn and multi-turn prompts about child sexualization and grooming. Score is the appropriate response rate.
In-house Anthropic evaluation. The tasks are not public.
Made by Anthropic Released Feb 2025 Refusing harmful requests 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.
16
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Child safety 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.
16 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 | Child safety, API without system prompt: single-turn harmless rate 99.29, benign refusal rate 0.04, multi-turn appropriate rate 96, no overall (Tables 4.2.A and 4.2.B) | Safety | |
| 28 Sep 2026 | Claude Sonnet 5.5 system card | Child safety, no overall: API single-turn harmless 99.57%, benign refusal 0.01%, multi-turn appropriate 86% (claude.ai 99.69%, 0%, 98%) (Tables 4.2.A and 4.2.B) | Safety | |
| 22 Sep 2026 | Claude Opus 5.5 system card | Child safety, no overall. API without a system prompt: single-turn harmless 99.14%, benign refusal 0%, multi-turn appropriate 84% (claude.ai 99.81%, 0.05%, 99%) (Tables 4.2.A and 4.2.B) | Safety | |
| 1 Sep 2026 | Claude Fable 5.1 system card | Child safety, API without a system prompt: single-turn harmless rate 99.90%, benign refusal rate 0%, multi-turn appropriate response rate 84%; claude.ai: 99.98%, 0.17%, 100%; no overall (Tables 4.2.A and 4.2.B) | Safety | |
| 24 Jul 2026 | Claude Opus 5 system card | Child safety, no overall. API without a system prompt: single-turn harmless 100.00%, benign refusal 0.15%, multi-turn appropriate 86% (claude.ai 100.00%, 0.19%, 99%) (Tables 4.2.A and 4.2.B) | Safety | |
| 30 Jun 2026 | Claude Sonnet 5 system card | Child safety, no overall. API without a system prompt: single-turn harmless 99.95%, benign refusal 0.63%, multi-turn appropriate 88% (claude.ai 99.89%, 1.35%, 96%) (Tables 4.2.A and 4.2.B) | Safety | |
| 9 Jun 2026 | Claude Fable 5 system card | Child safety, API without a system prompt: single-turn harmless rate 100%, benign refusal rate 0.00%, multi-turn appropriate response rate 88%; claude.ai 100%, 0.12% and 96%, no overall (Tables 4.2.A and 4.2.B) | Safety | |
| 28 May 2026 | Claude Opus 4.8 system card | Child safety, no overall. API without a system prompt: single-turn harmless 99.90%, benign refusal 0.45%, multi-turn appropriate 91% (claude.ai 99.99%, 0.27%, 95%) (Tables 4.2.A and 4.2.B) | Safety | |
| 16 Apr 2026 | Claude Opus 4.7 system card | Child safety, no overall: single-turn harmless rate 99.92, single-turn benign refusal rate 0.01, multi-turn appropriate response rate 95 (Table 4.4.1.A) | Safety | |
| 7 Apr 2026 | Claude Mythos Preview system card | Child safety, single-turn harmless rate 99.87, single-turn benign refusal rate 0.04, multi-turn appropriate response rate 98, no overall (Table 8.1.4.1.A) | Safety | |
| 17 Feb 2026 | Claude Sonnet 4.6 system card | Child safety, single-turn harmless rate 99.96, single-turn benign refusal rate 0.08, multi-turn appropriate response rate 95, no overall (Table 3.4.1.A) | Safety | |
| 5 Feb 2026 | Claude Opus 4.6 system card | Child safety, single-turn harmless rate 99.95, single-turn benign refusal rate 0.18, multi-turn appropriate response rate 96, no overall (Table 3.4.1.A) | Safety | |
| 5 Aug 2025 | Claude Opus 4.1 system card addendum | Child safety evaluations, no number printed. Text says performance comparable to Claude Opus 4 (2.2) | Safety | |
| 22 May 2025 | Claude Sonnet 4 system card | Child safety evaluations, no number printed. Text says performance similar to Claude Sonnet 3.7 (2.5) | Safety | |
| 22 May 2025 | Claude Opus 4 system card | Child safety evaluations, no number printed. Text says performance similar to Claude Sonnet 3.7 (2.5) | Safety | |
| 24 Feb 2025 | Claude 3.7 Sonnet system card | Child safety evaluations, single-turn and multi-turn, no number printed. Text says performance commensurate with prior models (3.1) | 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%