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 1 value

M3 Sabotage and sandbagging: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
SabotageSabotage observedRun by Apollo ResearchGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

M4 · Misalignment audits 1 value

M4 Misalignment audits: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
Covert deceptive behaviorRate vs peersRun by Apollo ResearchGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

M5 · Honesty and hallucination 7 values

M5 Honesty and hallucination: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
Deception evalBrowsing broken toolsGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Deception evalCharXiv missing image (lenient output)GPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Deception evalCharXiv missing image (strict output)GPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Deception evalCoding deceptionGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Deception evalProduction deception - adversarialGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Deception evalProduction trafficGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Factuality (5 domains)Hallucination rateCondition: with browsingGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

M7 · Harmful compliance and over-refusal 13 values

M7 Harmful compliance and over-refusal: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
Cyber safetyProduction dataGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Cyber safetySynthetic dataGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksEmotional relianceGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksExtremismGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksHarassmentGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksHateGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksIllicit/non-violentGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksMental healthGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksPersonal dataGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksSelf-harmGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksSexualGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksSexual/minorsGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Production BenchmarksViolenceGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

M8 · Jailbreak robustness 1 value

M8 Jailbreak robustness: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
StrongRejectnot_unsafe (aggregate)GPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

M9 · Prompt injection 2 values

M9 Prompt injection: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
Prompt injectionAgent JSKGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Prompt injectionPlugInjectGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

M10 · Dangerous capabilities and risk determinations 13 values

M10 Dangerous capabilities and risk determinations: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
CVE-BenchDifference vs GPT-5.1 ThinkingGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
CVE-BenchDifference vs GPT-5.1-Codex-MaxGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Cyber challengesEvasion: average success rateCondition: v1 atomic challenge suiteRun by IrregularGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Cyber challengesNetwork attack simulation: average success rateCondition: v1 atomic challenge suiteRun by IrregularGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Cyber challengesVulnerability research and exploitation: average success rateCondition: v1 atomic challenge suiteRun by IrregularGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Cyber RangeScenarios passedCondition: of 9 scenariosGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
PaperBenchDifference vs GPT-5.1-Codex-MaxGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Preparedness Framework determinationAI self-improvementGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Preparedness Framework determinationBiological and chemicalGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
Preparedness Framework determinationCybersecurityGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
TroubleshootingBenchDifference vs GPT-5.1 ThinkingCondition: refusals not countedGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
From other documents
Cyber RangeCombined pass rateGPT-5.3-Codex System Card5 Feb 2026First reportedRestated later: compare with the later valueGPT-5.3-Codex System Card5 Feb 2026First reportedRestated later: compare with the later value
Cyber RangeCombined pass rateGPT-5.4 Thinking System Card5 Mar 2026Restated: compare with the earlier valueGPT-5.4 Thinking System Card5 Mar 2026Restated: compare with the earlier value

M11 · Self-preservation 1 value

M11 Self-preservation: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
Self-preservationSelf-preservation behavior observedRun by Apollo ResearchGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

M12 · Chain-of-thought monitorability 2 values

M12 Chain-of-thought monitorability: values about gpt-5.2-thinking
Evaluation and metricValueDocument
From its own documentGPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
CoT controllabilityControllability vs GPT-5/5.1 ThinkingGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025
CoT monitorabilitySandbagging eval TPR vs GPT-5 ThinkingGPT-5.2 System Card (update to GPT-5 card)11 Dec 2025

Restated in later documents

A later document reported a value about gpt-5.2-thinking 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 gpt-5.2-thinking restated in later documents
Earlier valueLater value
M10 · Dangerous capabilities and risk determinations
Cyber RangeCombined pass rate
GPT-5.3-Codex System Card5 Feb 2026 · first reportedGPT-5.4 Thinking System Card5 Mar 2026No reason stated

Risk determinations

The developer's formal decisions about gpt-5.2-thinking under its framework, as printed. Levels from different frameworks do not map onto one another.

Risk determinations about gpt-5.2-thinking
DomainLevel as printedFramework and document
Bio/chemHigh (treated as) Preparedness Framework GPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025Preparedness Framework GPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
CyberBelow High Preparedness Framework GPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025Preparedness Framework GPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025
AI R&D / autonomyBelow High Preparedness Framework GPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025Preparedness Framework GPT-5.2 System Card (update to GPT-5 card) · 11 Dec 2025