OpenAI Unveils Enhanced Cybersecurity Model and Initiatives OpenAI is escalating its push into defensive cybersecurity, unveiling a more powerful AI model and a suite of programs aimed at moving the field from simply finding software flaws to fixing them at scale.

On June 22, OpenAI formally expanded its Daybreak initiative, describing it as a bid to “democratize patching vulnerable software at machine speed.” The company said it has already used its models to discover and generate patches for critical bugs in major browsers, network infrastructure, and operating systems like FreeBSD and the Linux kernel.

As part of this, OpenAI launched the full version of GPT‑5.5‑Cyber, following an earlier, more restricted preview. The model is available only to “trusted defenders” under a limited release and now sets “new state-of-the-art performance on CyberGym, reaching 85.6% compared with 81.8% for GPT‑5.5.” Axios, citing OpenAI’s blog, reports that the updated GPT‑5.5‑Cyber is “more permissive and more capable for advanced, authorized cybersecurity work,” able to analyze large codebases, validate likely vulnerabilities, and develop and test patches.

OpenAI also announced the Daybreak Cyber Partner Program, which lets vetted security vendors embed GPT‑5.5‑Cyber and related tools into their products, extending use beyond systems they directly own or manage. In parallel, it updated its Codex Security plugin to streamline discovery and patching and to help prevent new vulnerabilities from reaching production environments.

A major pillar of Daybreak is “Patch the Planet: a Daybreak initiative to support open source maintainers,” built with security firm Trail of Bits. OpenAI says it is “pairing AI-assisted security research using our most cyber-capable models with expert human review to not only identify vulnerabilities, but help patch them.” Security engineers triage findings before they reach maintainers, develop fixes, and coordinate disclosure, with initial participants including cURL, the Go project, Python, and key cryptography and networking libraries.

These moves land amid intense policy scrutiny. Axios notes that AI developers face a “difficult balancing act: getting powerful cyber capabilities into the hands of legitimate defenders and researchers while limiting opportunities for malicious use.” That debate is unfolding in parallel to regulatory discussions around rival Anthropic’s models, where observers argue U.S. policy is shifting toward “benchmark based tests of model failure” rather than assuming jailbreaks can be completely eliminated.

Overall, OpenAI’s Daybreak agenda reflects a bet that the new bottleneck in cybersecurity is no longer finding vulnerabilities, but patching them fast—and that carefully governed frontier models can help defenders keep pace.

Continue reading https://foxvector.com/stories/019ef1a3-2f60-14f6-7019-000a2e2f188c