AI models reportedly refuse to help OpenCode fight a sustained cyberattack

AI tools OpenCode was using for defense began refusing assistance during a sustained cyberattack, Andrew Curran reported on Oct. 18. Curran said analyzing the attacks triggered the models’ safety guardrails, and compared the problem with one he said Hugging Face had faced.

In the post Curran quoted, dax (@thdxr) described frustration while investigating the incident: “they did quite a bit and then started refusing.” The same post said access to “daybreak blue” had made the situation better.

Earlier on Oct. 18, dax described nonstop attempts to attack their systems over the preceding 48 hours. The post said the pressure was hardening the systems, but that they needed agents running around the clock because the attackers had them too.