AI-assisted coding platform OpenCode has announced a serious security incident that occurred on the night of October 17, 2026, in which linked API keys of a portion of its users were stolen and abused by attackers.

The incident stemmed from what OpenCode identified as a large scale distillation attack. This attack directly affected approximately 1.4% of all user accounts on OpenCode Console, specifically individuals and organizations that had linked paid accounts with third-party artificial intelligence model providers.

Nature of the Incident and Credential Theft Mechanism

According to an official announcement released on the morning of October 18, 2026, Vietnam time, OpenCode directly emailed warnings to all affected accounts. Customers who did not receive an email were confirmed to be safe from this attack.

The attackers’ primary goal in the incident was not merely reading source code or project data stored on the platform. Instead, the intruders focused on seizing API keys stored in the system to authorize service calls to major model developers, including Anthropic, OpenAI, and Google.

Many affected developers reported waking up to find their resource usage bills skyrocketing abnormally. The stolen API keys were fed into automated pipelines designed to extract model knowledge or execute resource-intensive tasks, leaving victims to shoulder the entire cost. OpenCode is currently working closely with security teams at Google, OpenAI, and Anthropic to revoke exposed credentials and prevent ongoing fraudulent activity.

Security Risks from Intermediary Model Orchestration Platforms

The incident at OpenCode raises a difficult question about current AI coding tool integration models. When users delegate full management of API keys to centralized dashboards for the convenience of using multiple models, security weaknesses in the intermediary platform turn directly into billing vulnerabilities for the users themselves.

To prevent similar risks, development teams are advised to immediately revoke API keys configured on OpenCode Console, establish hard spend limits within each model provider’s management dashboard, and switch to short-lived credentials with least-privilege scoping rather than granting broad, persistent access.