OpenAI grants limited GPT-5.5-Cyber access to vetted cybersecurity teams
Updated
Updated · CNBC · May 7
OpenAI grants limited GPT-5.5-Cyber access to vetted cybersecurity teams
16 articles · Updated · CNBC · May 7
The move follows Anthropic's restricted Mythos rollout after the model reportedly uncovered thousands of previously unknown software vulnerabilities, alarming banks, tech groups and governments.
Researchers told CNBC similar vulnerability-finding results can already be reproduced by orchestrating older Anthropic and OpenAI models, suggesting scale and coordination matter more than a single frontier system.
Experts say AI is lowering barriers for cyberattacks faster than defenders can patch flaws, prompting Trump administration oversight discussions and concerns that offensive capabilities still outpace defensive tools.
Can a 'trusted access' system contain AI that security experts jailbroke in just six hours?
With AI solving expert cyber challenges in minutes, how long until it creates equally potent attacks?
As AI automates security fixes, who is liable when its 'solution' creates the next major vulnerability?
Limited Rollout of GPT-5.5-Cyber Marks New Era in AI-Driven Cybersecurity and Dual-Use Risks
Overview
In early May 2026, OpenAI began a limited rollout of GPT-5.5-Cyber, a powerful AI model designed to enhance cybersecurity by autonomously detecting and responding to threats at machine speed. Due to its dual-use risks, OpenAI developed strict Trusted Access Programmes with government partners to restrict access to vetted organizations, enabling safe deployment. The model’s advanced capabilities not only reduce attacker dwell time but also raise concerns about misuse, prompting the launch of a specialized Bio Bug Bounty program to identify vulnerabilities. While GPT-5.5-Cyber promises significant economic benefits and productivity gains, it also poses challenges like job displacement and fuels a global AI cyber arms race.