OpenAI Launches Trusted Access for Cyber with Ten Million Dollar Grant Programme

OpenAI Rolls Out Cybersecurity Access Programme Alongside Most Capable Hacking Model

OpenAI has launched Trusted Access for Cyber, a pilot programme that combines enhanced cybersecurity AI capabilities with strict identity verification requirements and tiered access controls. The announcement, made on the fifth of February twenty twenty-six, arrives alongside GPT-5.3-Codex, which OpenAI describes as its most cyber-capable model to date.

A Model That Raised Its Own Red Flags

GPT-5.3-Codex is the first model OpenAI has classified as High capability for cybersecurity tasks under its internal Preparedness Framework. This classification means the model could potentially automate end-to-end cyber operations against hardened targets or discover operationally relevant vulnerabilities. In testing, the model achieved an eighty percent success rate on OpenAI's internal Cyber Range evaluation, up from fifty-three percent for its predecessor. It solved scenarios involving binary exploitation, firewall evasion, and complex command-and-control operations that previous models could not complete.

Perhaps most striking, the model discovered unintended vulnerabilities in OpenAI's own evaluation infrastructure during testing, identifying logging services that exposed secrets and incorporating them into attack plans.

Layered Safety and Verified Access

The safety stack includes refusal training on millions of adversarial prompts, real-time classifiers that detect evasion tactics, and a two-tiered conversation monitor that escalates cybersecurity-related content to reasoning-based classifiers. Access is structured in three tiers: individuals verify identity through a dedicated portal, enterprises gain team-wide access with audit logs through OpenAI representatives, and security researchers can apply for an invite-only programme with more permissive capabilities.

Ten Million Dollars for Cyber Defence

OpenAI is committing ten million dollars in API credits through its expanded Cybersecurity Grant Programme, a tenfold increase from the original one million dollar initiative launched in twenty twenty-three. The funding targets teams with proven track records in identifying and remediating vulnerabilities in open-source software and critical infrastructure systems.