OpenAI has launched Daybreak, an AI-powered cybersecurity program built on GPT-5.5 and Codex Security, taking direct aim at Anthropic's rival Project Glasswing.
Daybreak Targets Software Defense
The company unveiled the initiative on May 11, 2026. Daybreak is pitched as a way to bake cyber defense into software from day one rather than patching holes after the fact.
It builds on the April release of GPT-5.4-Cyber, which OpenAI credits with helping fix more than 3,000 vulnerabilities so far. The new program leans on Codex Security to construct an editable threat model straight from a company's repository.
OpenAI is offering three model tiers under the Daybreak umbrella.
Standard GPT-5.5 covers general-purpose work, while GPT-5.5 with Trusted Access for Cyber handles verified defensive tasks such as code review, vulnerability triage and patch validation. GPT-5.5-Cyber, the most permissive tier, sits in limited preview for red teaming and penetration testing.
Also Read: Sui Rallies 37% As Nasdaq Firm Locks Up 2.7% Of Supply
Industry Partners And Competitive Stakes
The launch arrives with a deep partner roster. OpenAI says it is already working with Cloudflare, Cisco, CrowdStrike, Palo Alto Networks, Oracle, Akamai, Fortinet and SentinelOne, among others.
OpenAI chief executive Sam Altman said the company wants to work with "as many companies as possible" to keep their software secure.
The pitch matters because Anthropic's Glasswing program, powered by the unreleased Claude Mythos Preview model, has already signed up Apple, Microsoft, Google and Amazon.
Glasswing has delivered early wins.
Mozilla disclosed in April that Mythos helped it find and patch 271 vulnerabilities in the latest Firefox release. Pricing for Daybreak is not yet listed, and OpenAI is routing interested organizations through a scan request or sales contact.
The move follows a steady cadence of cybersecurity launches from OpenAI in recent months. The company introduced Aardvark, a GPT-5-powered autonomous defense agent, on Oct. 29, 2025, then unveiled GPT-5.5-Cyber on May 7 before bundling its tools into Daybreak this week. Aardvark caught 92% of known and synthetic flaws in benchmark testing and has earned 10 CVE identifiers through open-source disclosures.
Read Next: Tom Lee Calls Crypto Spring As Bitmine Stakes $11.1B In ETH