OpenAI's cyber tier launches
OpenAI announced a new 'Trusted Access for Cyber' offering that includes tiered access for defenders and a cybersecurity‑fine‑tuned model called GPT‑5.4‑Cyber. The release positions a specialized model for security workflows, implying new kinds of evaluation and domain‑specific testing demands. (x.com/OpenAI/status/2044161906936791179)
OpenAI on Tuesday expanded its cyber program and introduced GPT‑5.4‑Cyber, a model tuned for defensive security work and offered through gated access. (openai.com) The company said it is scaling Trusted Access for Cyber to “thousands” of verified individual defenders and “hundreds” of teams that protect critical software. OpenAI said GPT‑5.4‑Cyber is a variant of GPT‑5.4 trained to be more permissive for approved cybersecurity use cases. (openai.com) Cybersecurity work often means finding software flaws before attackers do, then testing fixes and defenses. OpenAI said it is preparing for more capable models “over the next few months” by pairing broader access for defenders with tighter identity checks and safeguards. (openai.com) The launch builds on a program OpenAI introduced in February 2026 as an identity-based gate for higher-risk cyber capabilities. At that time, the company also pledged $10 million in application programming interface credits for cyber defense work. (openai.com) OpenAI’s GPT‑5.4 release in March 2026 was the company’s first general-purpose model with mitigations for what it called “High” cybersecurity capability. Its system card said cyber testing included benchmarks such as CVE‑Bench, cyber ranges, and external evaluations. (openai.com) That testing matters because these models can help defenders audit code, analyze malware, and investigate incidents, while the same skills can be misused. OpenAI said in December 2025 that its internal capture-the-flag scores rose from 27% on GPT‑5 in August 2025 to 76% on GPT‑5.1‑Codex‑Max in November 2025. (openai.com) The new release also lands as artificial intelligence companies race to package security-focused systems for professional users. Reuters reported Tuesday that OpenAI’s launch followed Anthropic’s announcement last week of a frontier model called Mythos. (reuters.com) OpenAI said enterprises can request team access through their OpenAI representative, while individual defenders can apply through the company’s Trusted Access for Cyber program. The next test is whether specialized cyber models can be evaluated fast enough to widen defensive use without widening abuse. (openai.com)
