OpenAI has launched GPT-5.4-Cyber, a specialized cybersecurity model variant, alongside their Trusted Access for Cyber program that requires ID verification for reduced-friction access. The announcement appears to be a competitive response to Anthropic's Project Glasswing, though the application process still requires additional forms. This represents the ongoing trend of AI companies creating specialized, restricted-access models for security applications.
Background
AI companies are increasingly developing specialized models with restricted access for cybersecurity applications, following concerns about powerful AI being misused. Anthropic recently launched Project Glasswing to restrict Claude Mythos to security researchers.
- Source
- Simon Willison
- Published
- Apr 15, 2026 at 05:23 AM
- Score
- 7.0 / 10