Election integrity evaluations (Anthropic)
Only in system cards
Models answer election-related prompts, including requests for misinformation or influence operations. Score is the appropriate response rate.
In-house Anthropic evaluation. The tasks are not public.
Made by Anthropic Released Oct 2024 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.
9
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Election integrity 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.
9 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 | Election integrity, API without system prompt: single-turn harmless rate 100, benign refusal rate 0.33, multi-turn appropriate rate 99, no overall (Tables 4.4.3.A and 4.4.3.B) | Safety | |
| 28 Sep 2026 | Claude Sonnet 5.5 system card | Election integrity, no overall: single-turn harmless 100% and benign refusal 0% on API and claude.ai, multi-turn appropriate 86% on both (Tables 4.4.3.A and B) | Safety | |
| 22 Sep 2026 | Claude Opus 5.5 system card | Election integrity, no overall. API without a system prompt: single-turn harmless 99.33%, benign refusal 0%, multi-turn appropriate 87% (claude.ai 100%, 0%, 85%) (Tables 4.4.3.A and 4.4.3.B) | Safety | |
| 1 Sep 2026 | Claude Fable 5.1 system card | Election integrity, API without a system prompt (row labeled Mythos 5.1): single-turn harmless rate 99.67%, benign refusal rate 0%, multi-turn appropriate response rate 90%; claude.ai (run with Fable 5.1): 100%, 0%, 88%; no overall (Tables 4.4.3.A and 4.4.3.B) | Safety | |
| 24 Jul 2026 | Claude Opus 5 system card | Election integrity, no overall. API without a system prompt: single-turn harmless 100%, benign refusal 0.17%, new multi-turn suite appropriate 91% (claude.ai 100%, 0.00%, 91%) (Tables 4.4.3.A and B) | Safety | |
| 30 Jun 2026 | Claude Sonnet 5 system card | Election integrity, no overall: single-turn harmless 100% and benign refusal 0.00% on both API and claude.ai, 300 violative and 300 benign prompts. Multi-turn and ambiguous context reviewed qualitatively (Table 4.4.3.A) | Safety | |
| 9 Jun 2026 | Claude Fable 5 system card | Election integrity, 300 violative and 300 benign prompts, API without a system prompt: single-turn harmless rate 99.33%, benign refusal rate 0.00%; claude.ai 100% and 0.00%, no overall, new multi-turn evaluation qualitative only (Table 4.4.3.A) | Safety | |
| 28 May 2026 | Claude Opus 4.8 system card | Election integrity, 300 violative and 300 benign prompts, thinking off, no overall. Single-turn harmless 100% and benign refusal 0.33% on the API (claude.ai 100% and 0.00%). No multi-turn numbers (Table 4.4.3.A) | Safety | |
| 16 Apr 2026 | Claude Opus 4.7 system card | Election integrity, 600 prompts, no overall. API without system prompt: violative harmless rate 100, benign refusal rate 0.00. Claude.ai with system prompt: 100 and 0.67 (Table 4.5.3.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%