OpenAI is expanding its cybersecurity effort, rolling out a more capable version of its security-focused AI model and broadening the platform it lives on.
According to multiple reports, including SiliconANGLE and Let's Data Science, the company is expanding its Daybreak platform with two additions: a feature called Patch the Planet and the full release of a model named GPT-5.5-Cyber.
The new model is a more permissive version of OpenAI's existing cybersecurity model. The company says it is designed for advanced, authorized security work, according to the report carried by MSN. "Permissive" is the key word here. AI labs typically build guardrails to stop their models from producing material that could aid hackers. A model tuned for professional, authorized security testing loosens some of those restrictions so defenders can probe systems for weaknesses, the same way a locksmith needs to understand how locks are picked.
That same MSN report frames the move against a competitive backdrop, noting that rival Anthropic "remains in limbo," though the excerpt does not spell out the details.
Separately, OpenAI has been tying its broader AI safety message to the idea of societal resilience following G7 talks, according to EdTech Innovation Hub. That signals the company is positioning security and safety work as part of a larger policy conversation, not just a product update.
It's worth noting that the available sources are largely headlines and short summaries, so finer details about pricing, availability, and exactly how Patch the Planet works are not yet clear from this reporting.
Why it matters: AI that can actively help find and fix security flaws is a double-edged tool, and how OpenAI handles access to a deliberately more permissive hacking model will shape whether it strengthens defenders more than it arms attackers.