Open-source AI coding tool OpenCode confirmed a breach that affected about 1.4% of OpenCode Console users, potentially exposing some model API keys stored on the platform. According to ChainCatcher, the incident involved unauthorized access to a legacy system being decommissioned and was linked by OpenCode to a large-scale model-distillation attack.

Developers reported unusual charges on accounts used to access models from OpenAI, Anthropic and Google. Colin Parsons reported about $281 in abnormal OpenAI charges and $168 in Anthropic charges, while Preston Zeller said he lost about $1,700; another user reported roughly $400 in unusual Anthropic charges. Some users also said unusual activity led Google to restrict their accounts.

OpenCode advised affected users to revoke and replace their keys and review recent API activity. Anthropic said it had worked with OpenCode and revoked Claude API keys believed to be affected. OpenAI team member Tibo said the company was helping affected API users restore access and would refund credits abnormally consumed during the attack. OpenCode co-creator Dax Raad said the breach affected Console, its cloud service, and not the locally run tool; keys configured directly in the local tool are not sent to OpenCode’s servers. On October 18, Raad said attackers had repeatedly tried to breach the system over the previous 48 hours and that the team was strengthening its defenses.