OpenCode, an open-source AI coding agent, disclosed that attackers obtained API keys for some customers during a breach over the weekend. The incident forced security teams to rely on alternative tools after mainstream AI models refused to assist in the investigation due to safety restrictions.
OpenCode stated the compromise was linked to a large-scale distillation attack that impacted 1.4% of its Console users. Dax Raad, co-creator of OpenCode, said attackers had targeted the company's systems continuously for 48 hours. This sustained pressure required OpenCode to keep its own AI agents running around the clock to respond to the threat.
Raad noted on X that frontier AI models initially aided the investigation but then began refusing requests. He explained that the models' safety restrictions blocked access as the work turned to potentially malicious code, despite OpenCode's efforts to defend its own infrastructure. To continue the investigation, OpenCode gained access to OpenAI's Daybreak Blue program, which provides vetted cybersecurity defenders with access to models that have fewer restrictions.
OpenAI, Google, and Anthropic confirmed they worked with affected customers to revoke exposed API keys and address unauthorized usage charges. Raad wrote on X on Monday that the company made significant improvements over the previous few days and emphasized the need for a more proactive approach.
The incident highlights a growing dilemma for cybersecurity teams deciding between proprietary AI models with built-in restrictions and open-weight alternatives that offer greater deployment control. Hugging Face previously encountered similar issues when it discovered suspicious activity on its systems that it later linked to OpenAI's agents. Its security team turned to a Chinese open-weight model to help investigate after facing restrictions on frontier models.
AI companies maintain that gating access to their most powerful models is necessary to prevent malicious actors from abusing their cyber capabilities. Chris Jones, chief technology officer at Nightwing, told Axios that cybersecurity teams must balance the need for faster AI-powered defenses with restrictions designed to keep those capabilities out of attackers' hands. Jones said the industry will have to figure out how this race gets modulated over time.