OpenCode Console Breach: OpenAI, Anthropic and Google API Keys Potentially Compromised
OpenCode has confirmed a breach of its Console service, with an unauthorised party gaining access to a legacy system on the night of 16 October. The company believes third-party API keys stored by users in Console may have been exposed.
For context: Console is OpenCode's optional hosted workspace. Teams can drop their OpenAI, Anthropic, or Google API keys into it so colleagues can access those models without ever handling the key directly. Convenient, until someone else gets hold of it.
Around 1.4% of Console users are affected, specifically those with an external provider key linked to their account. OpenCode sent emails to everyone in that group on the morning of 17 October with a subject line that didn't bury the lede: "Action Required: Potential provider key breach." If you didn't get that email, you're apparently in the clear.
The breach is confined to Console. Keys added locally through OpenCode's CLI or desktop app sit in a SQLite database on the user's own machine and never touch OpenCode's servers. Co-founder Dax Raad was fairly direct about this on X, noting the incident involves "the OpenCode Console, not OpenCode itself."
That distinction matters practically, but the exposure still carries real consequences. API keys are bearer credentials. There's no handshake, no identity check. Whoever has a copy can start firing billable requests at your account immediately. Deleting the key from Console doesn't revoke it at the provider level, and Console's own budget limits won't catch usage happening outside its gateway.
The remediation steps are straightforward, if a bit tedious. Revoke the compromised key in your provider's dashboard and generate a new one. Update anything using the old key. Check your billing history back to at least 16 October for requests you don't recognise. OpenCode hasn't confirmed when the unauthorised access actually started, so cast that net wide. Going forward, spend limits and per-tool keys are sensible housekeeping regardless.
Some users aren't waiting around to tally the damage. Reports are already emerging of real financial losses. Colin Parsons posted on X that his OpenAI account was hit for $281 and Anthropic for $168. Preston Zeller put his losses at $1,700. A Reddit user claimed $400 of unauthorised Anthropic usage; another reported roughly €200 of Google usage before the project was suspended entirely.
OpenCode's own language is carefully hedged. The keys "may have been exposed" is doing a lot of work there, and the company hasn't confirmed that any key was definitively copied or used. Whether that's accurate caution or reputation management is hard to say without more detail, and OpenCode hasn't offered much.
What the company has said is that the incident was still ongoing as of 17 October, that it was working with Anthropic, OpenAI, and Google, and that the breach was "related to a large-scale distillation attack." That phrase needs unpacking. Distillation attacks involve scraping a model's outputs at scale, typically to train a competing model on the cheap. Anthropic noted back in September that the groups behind these campaigns often get access through stolen credentials, fake accounts funded with stolen cards, and harvested API keys. A stolen key has other uses too: there's an active market for reselling AI access tokens.
How a distillation attack connects to a Console breach specifically, OpenCode hasn't explained. It also hasn't named any suspects, described the legacy system involved, or said how many keys were actually at risk in absolute numbers.
Raad posted on 18 October that attackers had been "non stop trying to attack our systems for the past 48 hours." The upside, he suggested, is that everything is getting hardened as a result. Bit of a rough way to run a security audit.