OpenCode, the open-source AI coding agent, said on October 17, 2026 that attackers breached its hosted OpenCode Console and may have exposed AI provider API keys saved by 1.4% of Console users. The company linked the breach to a "large scale distillation attack" and said it was working with Anthropic, OpenAI and Google. All three companies have since said they took action on affected keys or accounts.
Key Facts
| Details | |
|---|---|
| Disclosure | OpenCode post on X, October 17, 17:56 UTC (edited at 18:08 UTC) |
| When it happened | "Last night," according to OpenCode, meaning the night of October 16 |
| Affected users | 1.4% of OpenCode Console users, a subset of those who had linked external providers |
| Notification | Affected users were emailed on the morning of October 17; OpenCode says anyone without an email was not affected |
| Status | OpenCode called the incident ongoing and has not published further technical details |
What Users Were Told
The OpenCode email, as shared by users and reported by Notebookcheck and IT之家, says someone gained unauthorized access to a legacy system that OpenCode was in the process of shutting down. It warns that API keys stored in users' workspaces, including OpenAI and Anthropic keys, may have been exposed. Users were asked to revoke and replace those keys, update any applications that use them, and check recent usage for activity they did not make.
In the Reddit thread cited by both outlets, one user reported $400 in unauthorized charges on an Anthropic account. Another said Google usage charges reached about €200 and that Google suspended the project for abuse. These figures come from individual users, not from OpenCode or the providers.
How the Providers Responded
| Company | Response | Source |
|---|---|---|
| Anthropic | Revoked Claude API keys it believed were affected and asked users who saved a key in OpenCode to create a new one | Lydia Hallie of Anthropic's Claude Code team, October 17 |
| Blocked exposed Gemini API keys; asked users to create a new key in AI Studio, delete the old one, review usage and contact Google Cloud Billing support about unauthorized charges | Evan Otero of Google DeepMind and Google AI Studio, October 17 | |
| OpenAI | Worked with OpenCode to restore service for affected users who are OpenAI API customers and refund credits drawn during the attack | Tibo (@thsottiaux) of OpenAI's Codex team, October 19 |
Background
OpenCode lets developers use models from several providers with their own API keys, and the Console is its hosted service for workspaces and keys. In a distillation attack, someone collects large numbers of responses from a strong model to train another model. Stolen API keys let the attacker run those queries on the key owner's bill.
What We Don't Know Yet
- How many accounts 1.4% of Console users represents. OpenCode has not given a number.
- Who carried out the attack, and how the attacker got into the legacy system.
- Whether Anthropic and Google will refund charges run up with stolen keys. OpenAI is the only provider that has said it is refunding credits.