OpenCode says frontier models refused to help analyze attacks during its incident

OpenCode says frontier AI models it was using for defense began refusing to help analyze the attacking agents during a sustained cyberattack on its systems, because the analysis triggered safety guardrails.

The company says Hugging Face hit the same problem, according to a post by Andrew Curran. OpenCode says it has since gained access to Daybreak Blue, which it describes as working better for the investigation.