Top News

OpenAI launches Daybreak to help companies spot cyber threats
12 May 2026

OpenAI has unveiled Daybreak, a cutting-edge artificial intelligence (AI) initiative aimed at bolstering cybersecurity.

The announcement was made by CEO Sam Altman on X. He said, "OpenAI is launching Daybreak, our effort to accelerate cyber defense and continuously secure software."

The launch comes just weeks after Anthropic's controversial AI model, Claude Mythos, which raised global cybersecurity concerns due to its potential hacking capabilities.

Daybreak's approach to cyber defense
Cybersecurity strategy

OpenAI's Daybreak is based on the principle that future cyberdefense should be resilient by design. The idea is to build stronger defenses in an era where AI is becoming increasingly sophisticated.

While AI can help identify vulnerabilities and validate fixes, these same capabilities could also be exploited by malicious actors.

To tackle this, OpenAI has integrated its existing AI models with Codex's agentic capabilities for cybersecurity tasks like secure code review, threat modeling, patch validation, and dependency risk analysis.

Daybreak vs Project Glasswing
Partnership potential

OpenAI is keen to collaborate with as many companies as possible through Daybreak. OpenAI is already working with the likes of Cloudflare, Oracle, and Cisco for the same.

Unlike Anthropic's Project Glasswing that uses only Claude Mythos, OpenAI's Daybreak employs different AI models like GPT-5.5 and GPT-5.5-Cyber. This makes it more versatile in tackling cybersecurity challenges across various sectors.

Model options and features
Functionality details

Daybreak is designed to help teams identify and mitigate vulnerabilities, automate detection and response, and determine appropriate access levels for different workflows.

OpenAI has provided three model options under Daybreak: GPT-5.5 (default model with standard safeguards), GPT-5.5 with Trusted Access for Cyber (for verified defensive work in authorized environments), and GPT-5.5-Cyber (most advanced option for specialized authorized workflows).