Krux
April 16, 2026
OpenAI Unlocks GPT‑5.4-Cyber for Vetted Security Teams
Published: April 16, 2026 at 12:30 AM
Updated: April 16, 2026 at 12:30 AM
100-word summary
OpenAI released GPT-5.4-Cyber on April 14, a variant trained to help defenders find bugs and reverse-engineer malware without triggering the safety blocks that normally stop the base model cold. Security researchers get access to capabilities like binary reverse engineering, previously off-limits. The catch? You need to pass verification tiers under OpenAI's expanded Trusted Access for Cyber program, which now covers thousands of individual defenders and hundreds of teams. The move directly counters Anthropic's Mythos and Glasswing initiatives. OpenAI is betting that the best defense against AI-powered attacks is giving white hats the same weapons first.
What happened
OpenAI released GPT-5.4-Cyber on April 14, a variant trained to help defenders find bugs and reverse-engineer malware without triggering the safety blocks that normally stop the base model cold. Security researchers get access to capabilities like binary reverse engineering, previously off-limits. The catch? You need to pass verification tiers under OpenAI's expanded Trusted Access for Cyber program, which now covers thousands of individual defenders and hundreds of teams. The move directly counters Anthropic's Mythos and Glasswing initiatives.
Why it matters
OpenAI is betting that the best defense against AI-powered attacks is giving white hats the same weapons first.