What happened
OpenAI released the full GPT-5.5-Cyber on June 22 and expanded its Daybreak security effort. The company said the model scored 85.6% on the CyberGym benchmark, beating GPT-5.5 and Anthropic's Mythos 5, and shifts emphasis from finding vulnerabilities to patching them.
Access is gated through Trusted Access for Cyber, with partners including Cisco, CrowdStrike, Cloudflare, Palo Alto Networks, and Fortinet. Government partners span Australia, Canada, France, Germany, Japan, the EU's ENISA, and the UK. An updated Codex Security plugin scanned more than 30 million commits across over 30,000 codebases.
Why it matters
Automated patching, not just detection, is where security teams feel the shortage most. If a model can propose and validate fixes, it could change the economics of vulnerability response.
The gated rollout, tied to the June 2 US executive order, also shows how offensive-capable models are being handled: released to vetted defenders rather than the open public.
MintedBrain take
Gated access is the right default for tooling that doubles as an attack aid. Defenders who qualify should still treat generated patches as proposals and keep human review in the loop before merging.
Discussion
Sign in to comment. Your account must be at least 1 day old.