OpenAI launches Daybreak to take on Anthropic's Mythos in cyber defence

"The platform's working method begins with threat modelling against a given repository, then identifies and tests vulnerabilities in an isolated environment, and finally proposes and validates fixes. OpenAI says the goal is to compress security analysis that used to take hours into minutes, with audit-ready evidence handed back into enterprise systems."

"Daybreak rests on three model variants, according to OpenAI. GPT-5.5 covers general-purpose use under standard safeguards. GPT-5.5 with Trusted Access for Cyber is reserved for verified defenders performing tasks such as secure code review, vulnerability triage, malware analysis, and patch validation. A third model, GPT-5.5-Cyber, is a more permissive variant for authorised red teaming, penetration testing, and controlled validation."

"Launch partners include Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, and Zscaler, all of whom are integrating Daybreak capabilities under OpenAI's Trusted Access for Cyber initiative. Access is being kept tightly controlled at launch; organisations are being asked to request scans or speak to OpenAI sales."

"The launch positions OpenAI directly against Anthropic's Mythos, which has spent the past few months dominating the conversation about AI-powered defence. Mythos has surfaced thousands of zero-day vulnerabilities across major operating systems and browsers, and Anthropic has kept it"

Daybreak is a cybersecurity initiative focused on finding software vulnerabilities, generating patches, and validating fixes inside enterprise codebases. The platform uses three GPT-5.5 model variants: GPT-5.5 for general-purpose use under standard safeguards, GPT-5.5 with Trusted Access for Cyber for verified defenders performing secure code review, vulnerability triage, malware analysis, and patch validation, and GPT-5.5-Cyber for authorized red teaming, penetration testing, and controlled validation. The workflow starts with threat modelling for a repository, then tests vulnerabilities in an isolated environment, and finally proposes and validates fixes. The stated aim is to reduce security analysis time from hours to minutes while returning audit-ready evidence to enterprise systems. Launch partners include major security and networking companies integrating Daybreak capabilities under Trusted Access for Cyber, with access tightly controlled at launch.

Read at TNW | Openai