OpenAI and Anthropic keys may have leaked from OpenCode

OpenCode, the company behind the coding agent of the same name, said on Oct. 17 in its official X account that a security breach reached 1.4% of OpenCode Console users. All of them received a notice by email.

The notice describes unauthorized access to a legacy system the company was preparing to retire, according to DEV Community. Third-party keys that each user stored in their Console workspace "may have been exposed," and the notice names OpenAI and Anthropic as examples. An API key is the credential an app uses to reach an AI model, and the account holder pays for that use, so anyone who obtains one can run up charges in the owner's name.

According to the same outlet, which reproduces the instructions in the email, OpenCode recommends that recipients replace those keys in each provider's dashboard, swap the key in the apps and continuous-integration secrets where it was used, and check usage and billing from Oct. 16 at the latest. The420.in describes the same notice.

Users who did not receive an email are not considered affected, the company says. OpenCode is working with Anthropic, OpenAI and Google on remediation, and the incident remains open. Keys that users store on their own machine with the `/connect` command do not appear in what is known of the notice.

The company links the breach to what it called a "large-scale distillation attack," a term that usually means training one model on another model's answers. It has not explained how that relates to the access to the legacy system.

Four details are still missing: how many accounts the 1.4% represents, how the attacker got in, how long the access lasted, and whether all the keys were taken or only some. Some users have posted unauthorized charges of hundreds of dollars, but those are individual reports: neither OpenCode nor the providers have tied them to the incident. There is no statement with technical details.