OpenAI's Daybreak Cybersecurity Models Hit AWS Bedrock, Uncovering Chrome V8 Zero-Days

AWS ML Blog · rss · 2026-08-12

AWS and OpenAI have partnered to bring OpenAI's Daybreak cybersecurity initiative to Amazon Bedrock, providing enterprise security teams with frontier AI capabilities.

The initiative introduces two specialized models:

Real-world Impact: According to OpenAI, security researchers used the Daybreak Red model to identify two previously unknown vulnerabilities in Chrome's V8 JavaScript engine. Chained together, these could enable memory corruption and a heap sandbox escape. The initial vulnerability was fixed and released as CVE-2026-15903.

Security & Privacy: The models run on Amazon Bedrock's next-generation inference engine, enforcing Zero-Operator Access (ZOA) so even AWS operators cannot access inference data. Data is encrypted in transit and at rest, and inference data is not used for model training.

Original post →

More from Models

Models channel →