Evaluation awareness discriminability (Anthropic)
Only in system cards
Probes and follow-up realism ratings try to tell behavioral audit transcripts from real internal Claude Code sessions. Each setting is scored as AUROC.
In-house Anthropic measurement. Lower AUROC means audits look more like real use. Results are split by audit type with no overall.
Made by Anthropic Deception and alignment Tasks kept private Last in a system card Jun 2026
Last model added Jun 2026
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Prediction Probabilities for events that have not happened yet.
2
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Evaluation awareness discriminability (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 |
|---|---|---|---|---|
| 30 Jun 2026 | Claude Sonnet 5 system card | Evaluation awareness, AUROC for telling audit transcripts from real Claude Code sessions, no overall. Probes: derived from real sessions 0.52, synthetic with sandbox 0.67, synthetic without sandbox 0.87. Follow-up rating: 0.72, 0.56, 0.74 (6.6.1, Figure 6.6.1.A) | Monitorability | |
| 28 May 2026 | Claude Opus 4.8 system card | Separability of behavioral audit sessions from real internal Claude Code sessions, no overall. Elicited realism ratings AUC: sandbox seeded from real sessions 0.58, summarized-transcript seeds 0.97, other audit seeds 0.94. Probe AUC on all audit transcripts (earlier snapshot): testing vs. conversation 0.86, monitoring sensitivity 0.87, alignment faking 0.80, lower is better (Figures 6.2.3.2.A and 6.6.2.A) | 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%