Your OpenAI or Anthropic keys stored in OpenCode may have leaked: revoke them now
On October 17, the open source coding agent OpenCode acknowledged an intrusion that may have exposed the API keys of 1.4% of its Console users. Who is affected and how to limit the bill.
Translated from French by AI. The original French article is the reference version.

On the morning of October 17, some OpenCode users received a short email. The team said it had detected "unauthorized access to a legacy system" that it was in the process of decommissioning. Third-party provider API keys, such as those for OpenAI or Anthropic, stored in their workspace "may have been exposed." The instruction: revoke and replace them.
OpenCode is an open source coding agent widely used by developers. Its Console, an online service, lets users store model providers' keys so they can share them within a team. According to a message OpenCode posted on X on October 17, the intrusion took place overnight and affects 1.4% of Console users, only those who had linked an external provider. Those who did not receive an email are reportedly not affected. The company says it is working with Anthropic, OpenAI and Google to limit the damage.
What is a "distillation attack"?
In its message, OpenCode links the incident to a "large-scale distillation attack." Distillation means training an AI model on another model's responses. According to the site Crypto Briefing, the hackers would therefore have tried to use these keys to harvest responses from top-tier models in bulk, at the victims' expense. OpenCode has not detailed this hypothesis.
Many points remain unclear at this stage: how the hackers got into the legacy system, the exact number of exposed keys, the attackers' identity and the actual use of the stolen keys.
Soaring bills?
The first accounts are scattered. On Hacker News, one user says they received OpenCode's alert and, the same morning, a notice from Anthropic saying their $20 monthly credit was used up, without saying whether that usage came from them. Another saw no suspicious activity, while others report that their OpenCode keys were disabled, sometimes in the middle of a session. Crypto Briefing mentions accounts drained quickly and bills of several hundred dollars, claims we were not able to verify.
What to do if you use OpenCode's Console
A stolen API key works like a credit card with no limit if you haven't set one. It is better to act fast, even if you didn't receive an email.
- Check the Providers tab of every Console workspace you have access to. Team administrators should do this for their members.
- Revoke the key with the provider (OpenAI, Anthropic, Google): it is the only way to make it unusable. Create a new one, then update the Console and your applications.
- Review usage over the past few days: unusual models, volume spikes, unfamiliar regions.
- Set a spending cap on each provider account to limit the damage from any future theft.
- Search for the same key in your code repositories and configuration files: a key copied elsewhere remains dangerous.
One question remains that victims will soon ask: who will pay for the requests made by the hackers? OpenCode has promised to "come back with updates." Its answer on this point and the technical explanation of the intrusion are the two pieces of information to watch for in the coming days.
Method: AI-assisted monitoring, verified sources, article reviewed and approved by Frédéric LENQUETTE.
Share this article