OpenAI is launching Dawn, an AI initiative centered on detecting and patching vulnerabilities earlier than attackers discover them. Dawn makes use of the Codex Safety AI agent that launched in March to create a menace mannequin primarily based on a company’s code and concentrate on potential assault paths, validate probably vulnerabilities, after which automate the detection of the upper danger ones.
Its launch comes simply over a month after rival Anthropic introduced Claude Mythos, a security-focused AI mannequin it claimed was too harmful to publicly launch and solely shared privately as part of its personal initiative, dubbed Undertaking Glasswing. Nonetheless, that didn’t cease at the very least just a few unauthorized events from getting entry.
Nonetheless, OpenAI has thus far lacked the same safety product. Like Glasswing, Dawn isn’t constructed on only one AI mannequin — OpenAI says “Dawn brings collectively probably the most succesful OpenAI fashions, Codex, and our safety companions.”
Dawn additionally entails specialised cyber fashions, together with GPT-5.5 with Trusted Entry for Cyber and GPT-5.5-Cyber, which started rolling out final week. OpenAI additionally says it’s working with its “trade and authorities companions” whereas it prepares to “deploy more and more extra cyber-capable fashions.”
