Distillation attack on OpenCode drains users' API keys for Anthropic, OpenAI, Google models
Attackers breached OpenCode Console to hijack API keys, using them for large-scale model distillation and draining user funds.Summary
Anomaly Innovations disclosed a breach of its OpenCode Console platform, where attackers exploited legacy system vulnerabilities to hijack API keys from approximately 1.4% of users. These stolen credentials provided access to high-end AI models from Anthropic, OpenAI, and Google.
The incident is described as a "large scale distillation attack," where attackers use stolen credentials to run massive queries against frontier models. This process allows them to collect input-output pairs to train or refine cheaper, separate models, effectively offloading the high compute costs onto the victims. Affected users reported rapid depletion of account balances, with some facing bills in the hundreds of dollars.
OpenCode is currently working with major AI providers on remediation and has advised all users to immediately rotate or revoke their API keys. The breach highlights growing security concerns regarding the storage of third-party credentials within AI coding platforms and agentic workflows.
(Source：Vector Wire)