OpenCode API key breach: 1.4% of Console users exposed via a legacy system slated for shutdown
OpenCode said on October 17, 2026 that unauthorized access to a legacy Console system, already scheduled for shutdown, may have exposed AI provider API keys stored by about 1.4% of Console users with linked providers. It called the incident a large-scale distillation attack and told affected users, notified by email, to revoke and replace their keys
The three lines
- Scope — ~1.4% of Console users with linked providers; only those emailed
- Entry — a legacy system due to be retired; method and dwell time undisclosed
- Action — revoke and reissue keys, update apps, check provider usage and bills
Key questions
- OpenCode breach API keys
- **Stored provider keys of about 1.4% of Console users with linked providers may have been exposed.** | Item | Detail | |---|---| | Disclosed | October 17, 2026 (X) | | Entry | Legacy Console system | | Keys | OpenAI, Anthropic and others | | Notice | Emails to affected users |
- What to do after OpenCode breach
- **Revoke, reissue, review.** | Step | Action | |---|---| | 1 | Revoke old keys in each provider console | | 2 | Issue new keys and update apps | | 3 | Review usage and billing | | 4 | Report unexpected charges |
- OpenCode distillation attack
- **OpenCode's label; no technical evidence was published.** Distillation means harvesting a model's answers at scale to train another model; stolen keys let attackers do that on victims' bills.
An AI coding tool's key vault has leaked. OpenCode, an open-source AI coding agent, said on October 17, 2026 that overnight someone had accessed "a legacy system that was already scheduled to be shut down." About 1.4% of Console users who had linked external AI providers may have had their stored keys exposed.
1. What is confirmed
| Item | Detail |
|---|---|
| Disclosure | October 17, 2026, on X |
| Timing | Night of October 16-17 |
| Entry point | Legacy Console system due for retirement |
| Affected | ~1.4% of Console users with linked providers |
| Exposed | Stored third-party AI provider keys |
| Notice | Every affected user emailed; no email means not affected |
| Partners | Anthropic, OpenAI, Google |
The 1.4% is a share of users who linked providers in the Console, not of all OpenCode users. Absolute numbers were not given.
2. Why stolen keys matter
| Attacker gains | Victim bears |
|---|---|
| Free access to top models | The bill |
| Bulk model outputs for distillation | Possible account suspension |
| Cover: logs show the victim | Possible link to abuse |
Reports say attackers fired repeated queries at high-end models; one outlet noted they "were not after passwords or credit card numbers." They wanted model answers.
3. What to do
| Step | Action |
|---|---|
| 1 | Revoke keys in OpenAI, Anthropic and Google consoles |
| 2 | Issue new keys; update apps, servers and CI |
| 3 | Check recent usage and billing |
| 4 | Report unexpected charges to the provider |
| 5 | Set spending caps and alerts |
The structural lesson: tools that pool many providers' keys in one place turn one breach into several, and the door here was a system already slated to be switched off. The same week, Schwab and JPMorgan began blocking AI agents that log in with customer credentials, a related worry about handing keys to intermediaries.
4. What remains unconfirmed
- Attack method, dwell time and key count; no post-mortem yet.
- Evidence for the "distillation" label.
- Who pays for fraudulent usage.