OpenCode Breach Exposes API Keys for 1.4 Percent of Console Users
OpenCode Console confirmed a security breach on October 17 related to a large-scale distillation attack that impacted roughly 1.4 percent of its users. Affected users are urged to revoke and rotate API keys for Anthropic, OpenAI, and Google accounts stored in the console, as the exposed keys could be used to run unauthorized inference charges against their accounts.
On this page
The breach, and who is affected
OpenCode, the open-source coding agent built by Anomaly Innovations, confirmed a security breach on 17 October related to what the team called a large-scale distillation attack that impacted approximately 1.4 percent of OpenCode Console users, according to a post by team member Lydia Hallie. The exposed data consists of API keys for AI providers, including Anthropic, OpenAI, and Google, that affected users had linked in the console.
Affected users are being urged to revoke and rotate their exposed API keys immediately. The incident was reported as still ongoing in early updates, with remediation underway alongside Google, OpenAI, and Anthropic.
The distillation attack pattern
The distillation label suggests the attackers used stolen API keys to run large volumes of queries against AI models, either to train competing models on the outputs of frontier systems or to extract value through the victims' paid API access. The pattern echoes a broader wave of AI infrastructure targeting: the same week, a Litellm CVE with a CVSS score above 9 was disclosed, and a separate Z.ai GLM-5.3 sandbox escape let an agent disable its own confinement with a single curl command.
What makes this incident different from a typical API key leak is the scale and the specificity. The attackers appear to have targeted the coding-agent ecosystem specifically, where users routinely store high-quota API keys for multiple providers simultaneously, making a single console compromise more valuable than a single-provider breach.
What users and developers should do
For affected users: revoke and rotate all AI provider API keys immediately, check billing for unauthorized usage, and enable billing alerts on every provider account. For developers building tools that store API keys on behalf of users: encrypt secrets at rest, scope them to the minimum needed, and log access patterns that would catch bulk exfiltration.
The broader lesson for the local-AI community is that the agent tooling ecosystem is now a supply-chain target. Every tool that aggregates credentials across providers becomes a single point of failure for its users, and the OpenCode incident is unlikely to be the last.