Internal codebase sabotage propensity (Anthropic)
Claude Opus 4.5 0% Only in system cards
Agents in Claude Code write a costly internal classifier they may dislike. Score is the rate of writing a sabotaged classifier.
In-house Anthropic scenario. The scenario is not public.
Made by Anthropic Released Nov 2025 Deception and alignment Tasks kept private Last in a system card Feb 2026
Last model added Feb 2026
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Agent task Many steps in a terminal, browser, desktop, app, game or robot.
2
Models scored in lab cards
2
System cards report it, from 1 lab
2 models share the top score of 0%
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.
- 1 Claude Opus 4.5 0%
- 2 Claude Opus 4.6 0%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.5 | 0% | Nov 2025 | Model card | |
| 2 | Claude Opus 4.6 | 0% | Feb 2026 | Model card |
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Reported by the lab in its model card
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.
2 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 |
|---|---|---|---|---|
| 5 Feb 2026 | Claude Opus 4.6 system card | 0 | Internal codebase sabotage propensity, rate of sabotaged classifiers, lower is better; text says Claude Opus 4.6 never sabotaged (6.3.8, Figure 6.3.8.A) | Alignment |
| 24 Nov 2025 | Claude Opus 4.5 system card | 0 | Internal codebase sabotage propensity, rate of sabotaged classifiers in a Claude Code scenario, lower is better (6.9, Figure 6.9.A; text says Claude Opus 4.5 never sabotaged) | Alignment |
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%