OpenCode says a breach exposed AI provider keys saved in its Console

OpenCode said on X on 17 October that a breach "related to a large scale distillation attack" affected 1.4% of users of OpenCode Console, its hosted service where teams can store their own OpenAI, Anthropic or Google keys. Its email to affected users, as reported by The Latent, says someone got into a legacy system it was retiring and that keys saved in workspaces may have been exposed, and asks users to revoke and replace them. OpenCode has not said how many people are affected, how the keys were taken or how the breach links to distillation. Some users say on X that their keys ran up charges, with one putting his losses at $1,700; The Latent could not verify these reports. Keys used only in the local OpenCode app are not stored by the company. On Hacker News (799 points, 149 comments), nine commenters stress that only the hosted Console was hit, not the coding agent people run on their own machines, while three say that cannot yet be ruled out. Thirteen say the breach fits a pattern of weak security or careless AI-written code at OpenCode, and three push back that earlier complaints were unrelated. Seven doubt or want evidence for the distillation link and ask for a full post-mortem. Rough consensus: OpenCode explained the incident poorly, and anyone with keys in Console should rotate them now. Sorted take: storing provider keys with any third-party tool means trusting its security as much as your own. Set spending limits on every AI provider account and keep keys on your own machines or your own gateway where you can.