OpenAI announced an expansion of its Daybreak initiative with new tools designed to help organizations discover, validate, and patch software vulnerabilities automatically. The company revealed that AI has fundamentally shifted cybersecurity dynamics, where defenders are now overwhelmed by vulnerability discovery speed rather than struggling to find them. GPT-5.5-Cyber represents their strongest model yet for authorized security work.

Codex Security received a major plugin update that integrates defensive workflows directly into development environments. The tool can scan entire codebases, build threat models, trace attack paths, and generate targeted patches for review. Since March, the system has analyzed over 30 million commits across 30,000 codebases, with humans validating more than 70,000 findings as fixed. It exports results to existing vulnerability management systems using SARIF files and Code QL queries.

The updated GPT-5.5-Cyber achieved 85.6% accuracy on Cyber Gym benchmarks, outperforming its predecessor's 81.8%. It also scored 39.5% on Exploit Gym versus 25.95% for GPT-5.5, demonstrating superior vulnerability-to-exploit conversion capabilities. The model handles long-horizon security tasks across complex repositories while maintaining general-purpose intelligence.

Patch the Planet, launched with Trail of Bits and HackerOne, targets widely-used open-source projects including curl, Go, Python, and pyca/cryptography. Over 30 projects have committed to participate in the initiative, which aims to democratize frontier cyber capabilities through trusted access and human oversight. OpenAI's approach distributes powerful defensive tools beyond major tech companies to protect critical infrastructure.