OpenAI Launches GPT-5.5-Cyber: A Frontier Model Built for Cybersecurity

OpenAI is no longer content being a chatbot company — on June 22, it dropped its most powerful model straight into the cybersecurity trenches.

Three Key Takeaways

GPT-5.5-Cyber is OpenAI's strongest defensive cybersecurity model to date. It dominates the general-purpose GPT-5.5 across three major benchmarks: CyberGym at 85.6% (vs. 81.8%), ExploitGym at 39.5% (vs. 25.95%), and SEC-bench Pro at 69.8% (vs. 63.1%). The ExploitGym jump is the most dramatic — vulnerability exploitation detection capability improved by roughly 1.5x.

The model is not publicly available — it's gated to vetted defensive organizations only. OpenAI created the Trusted Access for Cyber program to control distribution. The initial partner list reads like a who's who of global cybersecurity: Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, and Zscaler. Companion tools include a Codex Security plugin and the Patch the Planet initiative supporting open-source maintainers.

Daybreak's ambition extends far beyond a single model. This expansion also introduced the Daybreak Cyber Partner Program, with over 25 security firms and several government agencies already on board. OpenAI's pitch is clear: turn frontier AI capabilities into measurable security risk reduction — not slide decks, but actual patches shipped.

WangDou's Take

There's an old saying in security: "Defenders have to plug every hole; attackers only need to find one." OpenAI is now offering a model that can scan massive codebases, trace call chains, verify vulnerability reachability, generate patches, and write reports — but only for the good guys. Sounds great, except that 39.5% on ExploitGym means it still whiffs on 60% of exploitation scenarios. And every "good guys only" tool in history — from Metasploit to Cobalt Strike — eventually ended up weaponized on the dark web. OpenAI's Daybreak has plenty of technical muscle. What it lacks is a gate that actually holds.

Source: The Hacker News · Cybersecurity News