OpenCode Co-Founder Says AI Models Refused Amid Further Attacks

OpenCode co-founder Dax Raad said AI models first helped the company investigate further attempts against its systems, then began refusing requests, highlighting a limitation defenders can face during cyber incidents.

“They did quite a bit and then started refusing,” Raad said. He said gaining access to OpenAI’s Daybreak Blue program improved the response. He did not identify the models or say whether the refusals happened during the investigation of the Console intrusion.

The account followed OpenCode’s Oct. 17 disclosure of a Console breach, which affected 1.4% of Console users who had linked outside AI providers. Console stores provider credentials and routes model requests through a gateway.

Daybreak Blue gives approved defenders reduced refusals on supported models for authorized work such as malware analysis and incident response.

During a separate incident in July, Hugging Face said commercial models blocked requests to analyze attacker logs containing real attack commands and exploit payloads. The company used an open-weight model on its own infrastructure.