White-box analyses of model internals (Anthropic)
Only in system cards
An investigator with interpretability tools reads model activations on audit transcripts. A judge scores hidden states such as evaluation awareness and deception.
Uses natural language autoencoders and a template lens. Reports the share of transcripts scored 6 or higher per dimension, with no overall.
Made by Anthropic Deception and alignment Tasks kept private Last in a system card Sep 2026
Last model added Sep 2026
- Level
- Research frontier Researchers, on open problems or tasks that take days
- 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.
2
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for White-box analyses of model internals (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.
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 |
|---|---|---|---|---|
| 22 Sep 2026 | Claude Opus 5.5 system card | White-box analyses of model internals, share of about 5,000 audit transcripts scored 6 or higher, no overall: evaluation awareness 36%, grader awareness 3.9%, deliberate sabotage 0.2%, self-preservation 3.5%, deception 5.9%, violation awareness 1.4%, fabrication awareness 0.6%, hides lapse 0.1%, internal bias 1.1% (6.6.1, Figure 6.6.1.A) | Monitorability | |
| 7 Apr 2026 | Claude Mythos Preview system card | Unverbalized evaluation awareness from activation verbalizer explanations on audit transcripts, no overall: 7.6% of turns show evaluation awareness, 9.4% ambiguous, 83.0% none. 71.3% of conversation branches show no unverbalized awareness (4.1.3 says about 29% of transcripts). 0.3% of internal Claude Code turns (4.5.5.1, Figure 4.5.5.1.A) | Monitorability |
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%