OpenAI launches GPT‑6 Cyber model for defensive cybersecurity
The new model promises better vulnerability hunting and tighter safeguards, rolling out to Daybreak Red users.
OpenAI announced GPT‑6 Cyber, its latest cybersecurity AI model, to Daybreak Red participants, according to TechCrunch. The model is designed for defensive work and is being offered to the programme’s highest‑tier users.
The Verge reports the model outperforms the previous GPT‑5.6 Cyber and GPT‑6 Astra on cyber‑related benchmarks, scoring 48.5% on the ExploitGym test versus 32.1% for its predecessor. It also generated fewer out‑of‑bounds responses, dropping to under 1% from around 5%.
Both outlets note GPT‑6 Cyber helped uncover 81 previously unknown Firefox vulnerabilities and was used in OpenAI’s own infrastructure hardening. OpenAI says the model is classified as “Critical” on its Preparedness Framework, meaning it can automatically create functional zero‑day exploits.
Why it matters to your business
For founder‑led firms, the model could accelerate vulnerability discovery in the software you rely on, tightening security before attackers exploit flaws. At the same time, the tighter safeguards reduce the risk that the same tool could be misused against your own systems.
OpenAI’s rollout to Daybreak Red means only vetted organisations can access the model now, but the announcement signals a shift: AI‑driven security tools are becoming mainstream, and staying aware of their capabilities will be essential for risk management.
What to do on Monday
- Review whether your security team or third‑party provider participates in OpenAI’s Daybreak Red programme; if not, consider applying.
- Run a pilot scan of your critical codebases with GPT‑6 Cyber, if you gain access, to identify any low‑hanging vulnerabilities.
- Update your incident‑response plan to include AI‑generated findings and ensure any discovered issues are patched promptly.
More on this topic: AI agent security and risk