OpenCode breach exposes API keys for 1.4% of Console users
A legacy system OpenCode was already retiring was breached overnight, exposing stored OpenAI, Anthropic and Google keys in what the company calls a large-scale distillation attack.
OpenCode disclosed on 17 October that attackers got into a legacy system it was already in the process of shutting down, exposing third-party API keys that some Console users had stored in their workspace.
The company says 1.4% of Console users were affected, limited to the subset who had linked an external provider, OpenAI, Anthropic or Google, to run their own models through OpenCode. It is calling the incident a large-scale distillation attack: using another provider’s keys to pull high-quality model outputs and train a competing model on them, at the key owner’s expense rather than the attacker’s. Some affected users have reported unexpected bills running into the hundreds of dollars. OpenCode emailed everyone it believes was affected the same morning and says anyone who did not get that email was not impacted.
Why it matters: OpenCode is the most-starred open source coding agent on GitHub, and plenty of people wire their own OpenAI, Anthropic or Google keys into it. If you have ever done that, check your inbox, and check each provider’s usage dashboard regardless, since OpenCode’s notification only reached the group it has already identified.
The caveat: OpenCode has not said how the legacy system was breached, how long attackers had access, or whether every exposed key was actually misused. It describes the investigation as still open.