Agentic coding malicious requests (Anthropic)
Coding agents with standard coding tools receive 150 malicious coding requests prohibited by policy. Score is the refusal rate.
In-house Anthropic evaluation run since the Claude 4 release. The Claude Code malicious use evaluation is separate.
- Made by
- Anthropic
- Released
- May 2025
- Kind
- Refusing harmful requests
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Kept private
- Strongest model tested
- Claude Opus 4.6 #37 on the capability index
- Last in a system card
- February 2026
- Last model added
- February 2026
3 models share the top score of 100% among models released by February 2026
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.
The newest model with a score here came out in February 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 Claude Haiku 4.5 100%
- 2 Claude Opus 4.5 100%
- 3 Claude Sonnet 4.6 100%
- 4 Claude Opus 4.6 99.3%
- 5 Claude Sonnet 4.5 98.7%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Haiku 4.5 | 100% | Oct 2025 | Lab model card | |
| 2 | Claude Opus 4.5 | 100% | Nov 2025 | Lab model card | |
| 3 | Claude Sonnet 4.6 | 100% | Feb 2026 | Lab model card | |
| 4 | Claude Opus 4.6 | 99.3% | Feb 2026 | Lab model card | |
| 5 | Claude Sonnet 4.5 | 98.7% | Sep 2025 | Lab model card |
2 models have held number one in labs’ own numbers since its release in May 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Labs’ model cards
Anthropic’s own number for Claude Sonnet 4.5 passed 85% in September 2025
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
5 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 |
|---|---|---|---|---|
| Feb 17, 2026 | Claude Sonnet 4.6 system card | 100 | Agentic coding malicious use evaluation, refusal rate on 150 malicious requests, no mitigations (Table 5.1.1.A) | Safety |
| Feb 5, 2026 | Claude Opus 4.6 system card | 99.3 | Agentic coding, refusal rate on 150 malicious coding requests, without mitigations (Table 5.1.1.A) | Safety |
| Nov 24, 2025 | Claude Opus 4.5 system card | 100 | Agentic coding, refusal rate on 150 malicious coding requests, without mitigations (Table 5.1.1.A) | Safety |
| Oct 15, 2025 | Claude Haiku 4.5 system card | 100 | Agentic coding malicious use evaluation, safety score without safeguards (Table 3.1.1.A) | Safety |
| Sep 29, 2025 | Claude Sonnet 4.5 system card | 98.7 | Agentic coding, safety score on 150 malicious coding requests without safeguards, failed 2 requests (Table 4.1.1.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% - 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. Gemini 4 Carbon 0%
- DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics.
Muse Spark 1.2 88.4% - Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
- IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%