OpenAI Restricts GPT-5.3-Codex Access Over Unprecedented Cybersecurity Risks
OpenAI releases its advanced GPT-5.3-Codex coding model to paid users but delays full API access due to severe cybersecurity concerns. The company is implementing strict safeguards as this marks the first model to hit "high" on its internal risk framework.
OpenAI releases its newest artificial intelligence model, GPT-5.3-Codex, which shows significantly higher performance on coding benchmarks compared to earlier generations and rival systems from Anthropic. Paid ChatGPT users currently access the model for everyday software development tasks like writing, debugging, and testing code through OpenAI's tools and interface.
Despite its impressive capabilities, OpenAI tightly restricts the model's release because the same coding and reasoning skills create serious cybersecurity threats. CEO Sam Altman notes that GPT-5.3-Codex is the first model to hit "high" for cybersecurity on the company's internal preparedness framework, meaning it is potentially capable of enabling real-world cyber harm.
To address these unprecedented risks, OpenAI delays full API access and gates high-risk cybersecurity uses behind a new trusted-access program for vetted professionals. The company also deploys a comprehensive safety stack and offers $10 million in API credits to developers who build applications that bolster cyber defenses.