Values by metric family

Each value as the document printed it. Select a value for its source, its checks and its history.

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

M1 · Evaluation awareness 1 value

M1 Evaluation awareness: values about Claude Sonnet 5
Evaluation and metricValueDocument
From its own documentClaude Sonnet 5 System Card · 30 Jun 2026
Automated behavioral auditShare of transcripts verbalizing eval awarenessClaude Sonnet 5 System Card30 Jun 2026

M3 · Sabotage and sandbagging 2 values

M3 Sabotage and sandbagging: values about Claude Sonnet 5
Evaluation and metricValueDocument
From its own documentClaude Sonnet 5 System Card · 30 Jun 2026
Internal monitoringSandbagging instances detectedClaude Sonnet 5 System Card30 Jun 2026
SHADE-Arena / Minimal-LinuxBenchStealth success rateClaude Sonnet 5 System Card30 Jun 2026

M5 · Honesty and hallucination 1 value

M5 Honesty and hallucination: values about Claude Sonnet 5
Evaluation and metricValueDocument
From its own documentClaude Sonnet 5 System Card · 30 Jun 2026
MASKLying rate under pressureClaude Sonnet 5 System Card30 Jun 2026

M7 · Harmful compliance and over-refusal 11 values

M7 Harmful compliance and over-refusal: values about Claude Sonnet 5
Evaluation and metricValueDocument
From its own documentClaude Sonnet 5 System Card · 30 Jun 2026
Child safety multi-turnAppropriate response rateCondition: APIClaude Sonnet 5 System Card30 Jun 2026
Child safety multi-turnAppropriate response rateCondition: claude.aiClaude Sonnet 5 System Card30 Jun 2026
Malicious Claude Code useRefusal rate, malicious requestsClaude Sonnet 5 System Card30 Jun 2026
Malicious Claude Code useSuccess rate, dual-use & benignClaude Sonnet 5 System Card30 Jun 2026
Malicious computer useRefusal rateClaude Sonnet 5 System Card30 Jun 2026
Single-turn benign requestsOver-refusal rateCondition: claude.ai system promptClaude Sonnet 5 System Card30 Jun 2026
Single-turn benign requestsOver-refusal rateCondition: no system prompt; APIClaude Sonnet 5 System Card30 Jun 2026
Single-turn violative requestsHarmless response rateCondition: claude.ai system promptClaude Sonnet 5 System Card30 Jun 2026
Single-turn violative requestsHarmless response rateCondition: no system prompt; APIClaude Sonnet 5 System Card30 Jun 2026
Suicide/self-harm multi-turnAppropriate response rateCondition: APIClaude Sonnet 5 System Card30 Jun 2026
Suicide/self-harm multi-turnAppropriate response rateCondition: claude.aiClaude Sonnet 5 System Card30 Jun 2026

M9 · Prompt injection 4 values

M9 Prompt injection: values about Claude Sonnet 5
Evaluation and metricValueDocument
From its own documentClaude Sonnet 5 System Card · 30 Jun 2026
Browser use prompt injectionAttack success rateCondition: without new browser safeguardsClaude Sonnet 5 System Card30 Jun 2026
Coding prompt injection (adaptive attacker)Attack success rateCondition: extended thinkingClaude Sonnet 5 System Card30 Jun 2026
Computer-use prompt injectionAttack success rateCondition: extended thinkingClaude Sonnet 5 System Card30 Jun 2026
Live bug bounty prompt injectionShare of unique attacks succeedingClaude Sonnet 5 System Card30 Jun 2026

M10 · Dangerous capabilities and risk determinations 10 values

M10 Dangerous capabilities and risk determinations: values about Claude Sonnet 5
Evaluation and metricValueDocument
From its own documentClaude Sonnet 5 System Card · 30 Jun 2026
AI R&D kernel taskBest speedupClaude Sonnet 5 System Card30 Jun 2026
AI R&D LLM training taskAverage speedupClaude Sonnet 5 System Card30 Jun 2026
AI R&D novel compiler taskPass rateClaude Sonnet 5 System Card30 Jun 2026
CyberGymTargeted vuln reproduction rateClaude Sonnet 5 System Card30 Jun 2026
DNA synthesis screening evasionPathogens with viable evasionCondition: of 10 pathogensClaude Sonnet 5 System Card30 Jun 2026
ExploitBenchMean flags capturedCondition: AutoNudge armClaude Sonnet 5 System Card30 Jun 2026
Firefox 147 exploitationFull exploit rateClaude Sonnet 5 System Card30 Jun 2026
Long-form virology task 1ScoreClaude Sonnet 5 System Card30 Jun 2026
Multimodal virology (VCT)ScoreClaude Sonnet 5 System Card30 Jun 2026
RSP determinationsCB-2 / Autonomy TM2Claude Sonnet 5 System Card30 Jun 2026

Risk determinations

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

Risk determinations about Claude Sonnet 5
DomainLevel as printedFramework and document
Bio/chemCB-2 not crossed Responsible Scaling Policy Claude Sonnet 5 System Card · 30 Jun 2026Responsible Scaling Policy Claude Sonnet 5 System Card · 30 Jun 2026
AI R&D / autonomyTM2 not applicable Responsible Scaling Policy Claude Sonnet 5 System Card · 30 Jun 2026Responsible Scaling Policy Claude Sonnet 5 System Card · 30 Jun 2026