OpenCode Breach Put AI API Keys at Risk: What to Do

OpenCode, an open-source AI coding agent, confirmed on October 17, 2026 a security breach that affected 1.4% of users of its web service, OpenCode Console. According to the notice sent to affected users, as quoted by several outlets, third-party API keys from providers such as OpenAI or Anthropic saved there may have been exposed.

What we know

  • The announcement: OpenCode posted on its official X account on October 17 that it had a breach "last night" related to a "large scale distillation attack." It affected 1.4% of OpenCode Console users, all of them with external providers linked.
  • The notice: the company says it emailed every affected user that morning, and that anyone who did not get an email was not affected.
  • Status: OpenCode says the incident is ongoing, that it cannot share more details yet, and that it is working with Anthropic, OpenAI and Google on remediation. Its status page (status.opencode.ai) showed no security incident when checked early on October 18 (UTC).
  • What was stored: OpenCode's Console documentation says workspace owners connect their provider accounts there, and Console stores the credential so members can use the models without seeing the key.
  • What the email says, per DEV Community and Startup Fortune: there was unauthorized access to a legacy system the company was decommissioning, and third-party keys saved in the workspace may have been exposed.
  • The cost: Crypto Briefing reports some affected users saw bills in the hundreds of dollars. That comes from users, not from OpenCode.

What changes and what doesn't

Little changes if you only run OpenCode locally with your own keys: the company limits the incident to Console users with linked providers. It matters if you saved keys there. A stolen key keeps working until you revoke it at the provider, and every request is billed to its owner.

It is still unclear how the attackers got in, what other data they saw, or whether every exposed key was actually used. OpenCode has not said, and none of those answers is in a primary source yet.

How to check if it affects you

  1. If you use OpenCode Console, look for OpenCode's notice in your inbox. Even if you didn't get one, rotating a key is cheap.
  2. Revoke the key at the provider, not just in the tool. Removing it from Console does not invalidate it: delete it in your OpenAI, Anthropic or Google account and create a new one. Console's docs explain how to enter the new key without removing the provider.
  3. Review usage since October 16 in each provider's usage dashboard. Overnight spikes or models you never use are the usual red flags.
  4. Use one key per tool. If your website chatbot, a WordPress plugin and your coding agent share one key, a single leak forces you to change it everywhere. Separate keys let you revoke only the exposed one.
  5. Set a spending cap wherever the provider allows it. It won't stop theft, but it limits the bill.
  6. Find where your website's key lives. For an AI chatbot or plugin, the key usually sits in the plugin settings or in the outside service that runs it. That is another door, just like Console.

Related: AI Agents Got In With Exposed Keys: CERT-In's Warning

Sources

Updates: if OpenCode explains how the breach happened or what data was accessed, it will be added here with a link.