Story perspectives

OpenAI launches GPT-5.5-Cyber preview for vetted cybersecurity defenders

5/9/2026

Summary

OpenAI Unveils Cyber

OpenAI began a limited preview of GPT-5.5-Cyber for vetted cybersecurity professionals protecting infrastructure.
The model uses the Trusted Access for Cyber framework to lower classifier refusals for vulnerability identification and patch validation.
OpenAI says GPT-5.5-Cyber adds no raw performance gain over GPT-5.5 but offers more permissive behavior for authorized defenders, requires users to enable Account Security starting June 1 2026, and partners with security vendors while extending Codex Security to open-source maintainers.

Summary

White House Reviews AI

The White House previewed GPT-5.5-Cyber to the Commerce Department’s AI Standards center and several Congressional committees while the International Monetary Fund warned misuse could destabilize the global economy.
U.S. officials consider executive action to vet advanced AI models before wider release, with Anthropic’s cyber policy head Rob Bair saying Mythos’ rollout gives defenders a months-long edge.
Silicon Valley leaders and the White House debate keeping permissive cybersecurity AI from adversaries.

History

Historical Context

OpenAI first withheld GPT-2 in 2019, citing fake-news risks, and turned profit-focused that year. After Microsoft secured exclusive GPT-3 licensing in 2020—GPT-3 has over 100 times more parameters than GPT-2—and Congress saw GPT-4’s tighter safety controls in 2023, OpenAI now limits GPT-5.5-Cyber to vetted cybersecurity teams, echoing the earlier cautious rollout.

1 of 3

OpenAI Unveils Cyber

OpenAI began a limited preview of GPT-5.5-Cyber for vetted cybersecurity professionals protecting infrastructure.
The model uses the Trusted Access for Cyber framework to lower classifier refusals for vulnerability identification and patch validation.
OpenAI says GPT-5.5-Cyber adds no raw performance gain over GPT-5.5 but offers more permissive behavior for authorized defenders, requires users to enable Account Security starting June 1 2026, and partners with security vendors while extending Codex Security to open-source maintainers.

1 / 3