1、GPT-5.3-Codex System CardOpenAIFebruary 5,20261Contents1Introduction32Baseline Model Safety Evaluations32.1Disallowed Content Evaluations.33Product-Specific Risk Mitigations43.1Agent sandbox.43.2Network access.54Model-Specific Risk Mitigations54.1Avoid data-destructive actions.54.1.1Risk description

2、.54.1.2Mitigation:Safety training.65Preparedness65.1Capabilities Assessment.65.1.1Biological and Chemical.65.1.1.1Tacit Knowledge and Troubleshooting.75.1.1.2ProtocolQA Open-Ended.85.1.1.3Multimodal Troubleshooting Virology.85.1.1.4TroubleshootingBench.95.1.2Cybersecurity.105.1.2.1Capture-the-flag(p

3、rofessional).125.1.2.2CVE-Bench.135.1.2.3Cyber Range.145.1.2.4External Evaluations by Irregular.175.1.3AI Self-Improvement.185.1.3.1Monorepo-Bench.185.1.3.2OpenAI-Proof Q&A.1915.1.4Research Category Update:Sandbagging.205.2Safeguards Assessment.215.2.1Cyber Safeguards.215.2.1.1Threat Model and Scena

4、rios.225.2.1.2Cyber Threat Taxonomy.225.2.1.3Safeguards.235.2.1.3.1Model Safety Training.245.2.1.3.2Conversation monitor.245.2.1.3.3Expert Red Teaming.255.2.1.3.4Actor Level Enforcement.275.2.1.3.5Trust-based access.275.2.1.4Security Controls.285.2.1.5Misalignment risks and internal deployment.285.2

5、.1.6Sufficiency of Risk Mitigation Measures.2921IntroductionGPT-5.3-Codex is the most capable agentic coding model to date,combining the frontier codingperformance of GPT-5.2-Codex with the reasoning and professional knowledge capabilities ofGPT-5.2.This enables it to take on long-running tasks that

6、 involve research,tool use,andcomplex execution.Much like a colleague,you can steer and interact with GPT-5.3-Codex whileits working,without losing context.Like other recent models,it is being treated as High capability on biology,and is being deployedwith the corresponding suite of safeguards we us