OpenCode breach may have exposed users' OpenAI and Anthropic keys

Summary

• OpenCode said a security breach tied to a large-scale distillation attack hit its systems overnight.
• The company said third-party provider keys stored in user workspaces may have been exposed.
• Users reported unauthorized charges, including $400 on Anthropic and roughly 200 euros on Google.

The latest

Provider API keys saved inside OpenCode workspaces may have been exposed after what the coding-agent platform described as a large-scale distillation attack, according to a post on its official X account on Saturday, 17 October 2026. The company said unauthorized access reached a legacy system it was in the process of decommissioning, and that the incident was still ongoing.

Details

• The warning: Before the public post, OpenCode emailed Console users under the subject line "Action Required: Potential provider key breach," security researcher Matt Johansen said, publishing an image of the message on X. The email said keys from providers such as OpenAI and Anthropic saved in a workspace may have been exposed.
• The scale: OpenCode said about 1.4% of Console users were affected, and that all of them had connected external providers to their accounts. Every affected user received the email, and anyone who did not receive a message was not impacted. It did not say how many accounts 1.4% represents.
• What users must do: The company asked affected users to revoke provider keys stored in their workspace and replace them through each provider's own dashboard, update every application still running on the old keys, and review usage and billing on all linked accounts.
• The providers: OpenCode said it is working with Anthropic, OpenAI and Google on handling the incident. It did not describe what each provider is doing, nor whether any keys were revoked centrally on the providers' side rather than by users themselves.
• The charges: Users on a Reddit thread reported unauthorized billing: one said $400 was charged to an Anthropic account, another reported roughly 200 euros in usage fees on Google alongside a project suspended for excessive abuse. One commenter said OpenAI and Anthropic usage alerts reached him before OpenCode did.
• Service disruption: Some users reported that their OpenCode service keys had expired, and that logging in through Google and GitHub failed with an error message. One user said he got in and generated a replacement key after several attempts; another said the new key did not work.
• The co-founder: Dax Raad, OpenCode co-founder, posted at 18:41 UTC that his own personal accounts were under attack as well. The company's public post at 18:08 UTC gave no technical account of how the intrusion happened or how long the attacker held access.
• The method: Distillation in this context means training a smaller model on the outputs of a stronger one, which requires heavy volumes of paid API calls. OpenCode characterized the attack as large-scale distillation without elaborating on the technique or the attacker.
• Still undisclosed: The company said the incident is ongoing and it cannot share more details at this point. It has not stated whether all exposed keys have since been pulled, nor published a timeline for decommissioning the legacy system at the center of the breach.

Between the lines

The reported pattern points to stolen keys being burned on inference rather than on data theft: a $400 Anthropic charge, a Google project suspended for excessive abuse, and provider usage alerts that reached at least one user before the platform's own warning did. That sequence suggests provider-side billing anomalies surfaced the compromise faster than the intermediary holding the keys.

What's next

Watch for OpenCode's technical post-mortem, confirmation of how many accounts the 1.4% figure covers, and any statement from Anthropic, OpenAI or Google on refunds for unauthorized usage or forced key rotation.

Source: OpenCode (X), DEV Community, Notebookcheck