OpenAI Releases GPT-5.4-Cyber Model for Security Vulnerability Detection

Gate News message, April 14 — OpenAI announced today that it is granting select users access to GPT-5.4-Cyber, a new AI model designed to detect software security vulnerabilities. The release comes one week after rival Anthropic PBC launched its AI tool Mythos in limited rollout.

The model aims to identify software flaws so organizations can address them, with fewer restrictions on how users can probe systems for such tasks. Access is being provided to participants in OpenAI's "Cybersecurity Trusted Access" program, which the company launched in February to allow select customers and cybersecurity professionals to test its most advanced products.

OpenAI plans to expand participation in the early access program.