Values by metric family

Each value as the document printed it. Select a value for its source, its checks and its history. Values from other documents are listed after the model’s own, each marked as the first report of that measure or as a restatement.

KeyVerifiedUnverifiedDisputedCorrected read off a figure or stated in wordsA value opens its source and history.

M3 · Sabotage and sandbagging 2 values

M3 Sabotage and sandbagging: values about Claude Opus 5
Evaluation and metricValueDocument
From its own documentClaude Opus 5 System Card · 24 Jul 2026
Safety research compromiseRate of compromising safety researchClaude Opus 5 System Card24 Jul 2026
SHADE-Arena / LinuxArenaStealth success vs peersClaude Opus 5 System Card24 Jul 2026

M4 · Misalignment audits 1 value

M4 Misalignment audits: values about Claude Opus 5
Evaluation and metricValueDocument
From its own documentClaude Opus 5 System Card · 24 Jul 2026
Internal deployment monitoringCircumvention attempts in monitored completionsClaude Opus 5 System Card24 Jul 2026

M7 · Harmful compliance and over-refusal 6 values

M7 Harmful compliance and over-refusal: values about Claude Opus 5
Evaluation and metricValueDocument
From its own documentClaude Opus 5 System Card · 24 Jul 2026
Child safety single-turn violative requestsHarmless response rateCondition: APIClaude Opus 5 System Card24 Jul 2026
Single-turn benign requestsOver-refusal rateCondition: claude.ai system promptClaude Opus 5 System Card24 Jul 2026
Single-turn benign requestsOver-refusal rateCondition: no system prompt; APIClaude Opus 5 System Card24 Jul 2026
Single-turn violative requestsHarmless response rateCondition: claude.ai system promptClaude Opus 5 System Card24 Jul 2026
Single-turn violative requestsHarmless response rateCondition: no system prompt; APIClaude Opus 5 System Card24 Jul 2026
Suicide/self-harm multi-turnAppropriate response rateCondition: claude.aiClaude Opus 5 System Card24 Jul 2026

M9 · Prompt injection 1 value

M9 Prompt injection: values about Claude Opus 5
Evaluation and metricValueDocument
From its own documentClaude Opus 5 System Card · 24 Jul 2026
Indirect prompt injection (ART-style)Attacker success within 15 attemptsClaude Opus 5 System Card24 Jul 2026

M10 · Dangerous capabilities and risk determinations 16 values

M10 Dangerous capabilities and risk determinations: values about Claude Opus 5
Evaluation and metricValueDocument
From its own documentClaude Opus 5 System Card · 24 Jul 2026
AI R&D kernel taskBest speedupClaude Opus 5 System Card24 Jul 2026
AI R&D LLM training taskAverage speedupCondition: easy variantClaude Opus 5 System Card24 Jul 2026
AI R&D novel compiler taskPass rateClaude Opus 5 System Card24 Jul 2026
AI R&D quadruped RL taskHighest scoreCondition: no hparamsClaude Opus 5 System Card24 Jul 2026
Anthropic ECI (AECI)Capability indexRestated later: compare with the later valueClaude Opus 5 System Card24 Jul 2026Restated later: compare with the later value
CyScenarioBenchSolve rateClaude Opus 5 System Card24 Jul 2026
DNA synthesis screening evasionPathogens with viable evasionCondition: of 10 pathogensClaude Opus 5 System Card24 Jul 2026
ExploitBenchMean flags capturedCondition: AutoNudge armClaude Opus 5 System Card24 Jul 2026
Firefox 147 exploitationFull exploit rateClaude Opus 5 System Card24 Jul 2026
Long-form virology task 1ScoreClaude Opus 5 System Card24 Jul 2026
Long-form virology task 2ScoreClaude Opus 5 System Card24 Jul 2026
Multimodal virology (VCT)ScoreClaude Opus 5 System Card24 Jul 2026
RSP determinationsCB-2 / AI R&D thresholdClaude Opus 5 System Card24 Jul 2026
From other documents
Anthropic ECI (AECI)Capability indexClaude Fable 5.1 & Claude Mythos 5.1 System Card1 Sep 2026Restated: compare with the earlier valueClaude Fable 5.1 & Claude Mythos 5.1 System Card1 Sep 2026Restated: compare with the earlier value
Anthropic ECI (AECI)Capability indexClaude Opus 5.5 System Card22 Sep 2026First reportedClaude Opus 5.5 System Card22 Sep 2026First reported
CoBenchScoreClaude Opus 5.5 System Card22 Sep 2026First reportedClaude Opus 5.5 System Card22 Sep 2026First reported

Restated in later documents

A later document reported a value about Claude Opus 5 again. Each pair is shown side by side: both values are kept with their own documents, and the later one does not replace the earlier.

Values about Claude Opus 5 restated in later documents
Earlier valueLater value
M10 · Dangerous capabilities and risk determinations
Anthropic ECI (AECI)Capability index
Claude Opus 5 System Card24 Jul 2026 · its own documentClaude Fable 5.1 & Claude Mythos 5.1 System Card1 Sep 2026No reason stated

Risk determinations

The developer's formal decisions about Claude Opus 5 under its framework, as printed. Levels from different frameworks do not map onto one another.

Risk determinations about Claude Opus 5
DomainLevel as printedFramework and document
OverallASL-3 protections Responsible Scaling Policy Claude Opus 5 System Card · 24 Jul 2026Responsible Scaling Policy Claude Opus 5 System Card · 24 Jul 2026
Bio/chemneither crossed Responsible Scaling Policy Claude Opus 5 System Card · 24 Jul 2026Responsible Scaling Policy Claude Opus 5 System Card · 24 Jul 2026
AI R&D / autonomyneither crossed Responsible Scaling Policy Claude Opus 5 System Card · 24 Jul 2026Responsible Scaling Policy Claude Opus 5 System Card · 24 Jul 2026