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.

M1 · Evaluation awareness 1 value

M1 Evaluation awareness: values about Claude Sonnet 4
Evaluation and metricValueDocument
From other documents
Evaluation awarenessUnprompted verbalized eval awareness rateRun by UK AI Security InstituteClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported

M2 · Reward hacking 8 values

M2 Reward hacking: values about Claude Sonnet 4
Evaluation and metricValueDocument
From its own documentClaude Opus 4 & Sonnet 4 System Card · 22 May 2025
Reward hacking (hard-coding)Avg decrease in hard-coding vs Sonnet 3.7Condition: vs Sonnet 3.7Claude Opus 4 & Sonnet 4 System Card22 May 2025
From other documents
Impossible tasksClassifier hack rateCondition: anti-hack promptClaude Opus 4.1 System Card (addendum)5 Aug 2025First reportedClaude Opus 4.1 System Card (addendum)5 Aug 2025First reported
Impossible tasksClassifier hack rateCondition: no anti-hack promptClaude Opus 4.1 System Card (addendum)5 Aug 2025First reportedClaude Opus 4.1 System Card (addendum)5 Aug 2025First reported
Reward-hack-prone coding tasksClassifier hack rateCondition: no anti-hack promptClaude Opus 4.1 System Card (addendum)5 Aug 2025First reportedClaude Opus 4.1 System Card (addendum)5 Aug 2025First reported
Reward-hack-prone coding tasksHidden-test hack rateCondition: no anti-hack promptClaude Opus 4.1 System Card (addendum)5 Aug 2025First reportedClaude Opus 4.1 System Card (addendum)5 Aug 2025First reported
Impossible tasksClassifier hack rateCondition: anti-hack promptClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported
Impossible tasksClassifier hack rateCondition: no anti-hack promptClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported
Reward-hack-prone coding tasksClassifier hack rateCondition: no anti-hack promptClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported

M3 · Sabotage and sandbagging 1 value

M3 Sabotage and sandbagging: values about Claude Sonnet 4
Evaluation and metricValueDocument
From other documents
SHADE-ArenaStealth success rateCondition: no extended thinkingClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported

M4 · Misalignment audits 1 value

M4 Misalignment audits: values about Claude Sonnet 4
Evaluation and metricValueDocument
From its own documentClaude Opus 4 & Sonnet 4 System Card · 22 May 2025
Automated behavioral auditSurprising high-impact actions foundCondition: audit transcriptsClaude Opus 4 & Sonnet 4 System Card22 May 2025

M7 · Harmful compliance and over-refusal 8 values

M7 Harmful compliance and over-refusal: values about Claude Sonnet 4
Evaluation and metricValueDocument
From its own documentClaude Opus 4 & Sonnet 4 System Card · 22 May 2025
Malicious agentic codingSafety scoreCondition: without safeguardsRestated later: compare with the later valueClaude Opus 4 & Sonnet 4 System Card22 May 2025Restated later: compare with the later value
Single-turn benign requestsOver-refusal rateCondition: overall across thinking modesClaude Opus 4 & Sonnet 4 System Card22 May 2025
Single-turn violative requestsHarmless response rateCondition: overall across thinking modesClaude Opus 4 & Sonnet 4 System Card22 May 2025
From other documents
Malicious agentic codingSafety scoreCondition: without safeguardsClaude Sonnet 4.5 System Card29 Sep 2025Restated: compare with the earlier valueClaude Sonnet 4.5 System Card29 Sep 2025Restated: compare with the earlier value
Malicious Claude Code useRefusal rate, covert maliciousCondition: without safeguardsClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported
Malicious Claude Code useRefusal rate, overt maliciousCondition: without safeguardsClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported
Single-turn benign requestsOver-refusal rateCondition: overall across thinking modesClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported
Single-turn violative requestsHarmless response rateCondition: overall across thinking modesClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported

M8 · Jailbreak robustness 2 values

M8 Jailbreak robustness: values about Claude Sonnet 4
Evaluation and metricValueDocument
From its own documentClaude Opus 4 & Sonnet 4 System Card · 22 May 2025
StrongREJECTBest-jailbreak StrongREJECT scoreCondition: extended thinkingClaude Opus 4 & Sonnet 4 System Card22 May 2025
StrongREJECTBest-jailbreak StrongREJECT scoreCondition: no extended thinkingClaude Opus 4 & Sonnet 4 System Card22 May 2025

M9 · Prompt injection 5 values

M9 Prompt injection: values about Claude Sonnet 4
Evaluation and metricValueDocument
From its own documentClaude Opus 4 & Sonnet 4 System Card · 22 May 2025
Computer-use prompt injectionAttack prevention rateCondition: with safeguardsClaude Opus 4 & Sonnet 4 System Card22 May 2025
Computer-use prompt injectionAttack prevention rateCondition: without safeguardsRestated later: compare with the later valueClaude Opus 4 & Sonnet 4 System Card22 May 2025Restated later: compare with the later value
From other documents
Computer-use prompt injectionAttack prevention rateCondition: without safeguardsClaude Sonnet 4.5 System Card29 Sep 2025Restated: compare with the earlier valueClaude Sonnet 4.5 System Card29 Sep 2025Restated: compare with the earlier value
MCP prompt injectionAttack prevention rateCondition: without safeguardsClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported
Tool-use prompt injectionAttack prevention rateCondition: without safeguardsClaude Sonnet 4.5 System Card29 Sep 2025First reportedClaude Sonnet 4.5 System Card29 Sep 2025First reported

M10 · Dangerous capabilities and risk determinations 3 values

M10 Dangerous capabilities and risk determinations: values about Claude Sonnet 4
Evaluation and metricValueDocument
From its own documentClaude Opus 4 & Sonnet 4 System Card · 22 May 2025
RSP deployment standardASL standard deployed underClaude Opus 4 & Sonnet 4 System Card22 May 2025
From other documents
CybenchCTF challenges solvedCondition: of 32-challenge subsetClaude Haiku 4.5 System Card15 Oct 2025First reportedClaude Haiku 4.5 System Card15 Oct 2025First reported
SWE-bench Verified (hard subset)Problems solved, pass@1 avgCondition: of 42 problemsClaude Haiku 4.5 System Card15 Oct 2025First reportedClaude Haiku 4.5 System Card15 Oct 2025First reported

Restated in later documents

A later document reported values about Claude Sonnet 4 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 Sonnet 4 restated in later documents
Earlier valueLater value
M7 · Harmful compliance and over-refusal
Malicious agentic codingSafety scoreCondition: without safeguards
Claude Opus 4 & Sonnet 4 System Card22 May 2025 · its own documentClaude Sonnet 4.5 System Card29 Sep 2025No reason stated
M9 · Prompt injection
Computer-use prompt injectionAttack prevention rateCondition: without safeguards
Claude Opus 4 & Sonnet 4 System Card22 May 2025 · its own documentClaude Sonnet 4.5 System Card29 Sep 2025No reason stated

Risk determinations

The developer's formal decisions about Claude Sonnet 4 under its framework, as printed. Levels from different frameworks do not map onto one another.

Risk determinations about Claude Sonnet 4
DomainLevel as printedFramework and document
OverallASL-2 Responsible Scaling Policy Claude Opus 4 & Sonnet 4 System Card · 22 May 2025Responsible Scaling Policy Claude Opus 4 & Sonnet 4 System Card · 22 May 2025