For: people who use OpenCode Console, the hosted service of the open-source AI coding agent OpenCode, especially those who linked API keys from external AI providers in a Console workspace. This article is based on OpenCode's official X post, with details from the company's email and statements as reported by news outlets, and does not cover attack techniques.
Developing: this article will be updated as the organisation publishes more
- Who is affected
people who use OpenCode Console, the hosted service of the open-source AI coding agent OpenCode, especially those who linked API keys from external AI providers in a Console workspace.
- Confirmed so farSee the table ↓
- Share of Console users impacted
- Who was impacted
- Notice to affected users
Listed as possibly affected
- External AI provider keys stored in Console workspaces
- Not yet confirmed
- How the attacker got in
- How many keys were exposed and how they were stored
- How far the keys were misused
1 more
- Conditions for refunds
- What to do nowSee the steps ↓
- Check whether you received OpenCode's email
- Revoke linked AI provider keys at each provider
- Check usage and billing from October 17
2 more
- Set limits at the provider before using the new key
- If you were charged, keep records and contact both companies
- Where to verify (source)All sources ↓
- OpenCode, official X account (@opencode, linked from opencode.ai), post of October 17, 2026
- Last checked
- As of October 31, 2026
Follow-up plans and what we checked
The October 17 post says: "The incident is ongoing so we cannot share more details at this point." As of October 31, no further official statement on this incident was found on the company's status page or in a web search.
What OpenCode Console users should do today
Check whether you received OpenCode's email
OpenCode says it emailed every affected user on the morning of October 17, 2026, and that users who did not get an email were not impacted. Search your inbox, including spam, for a message from OpenCode around that date.
Do not use links in emails or messages claiming to be about this incident. To check your account, open OpenCode Console from a bookmark or by typing the address yourself.
Revoke linked AI provider keys at each provider
If you got the email, revoke every external AI provider key you had stored in OpenCode Console. According to news reports, this is what OpenCode told affected users. Do it in the provider's own dashboard: deleting the key from OpenCode does not stop a copy someone already has.
If you did not get the email but had keys linked in Console and want to be sure, revoking and replacing them costs little. Update every place that used the old key so your own tools keep working.
Check usage and billing from October 17
In each provider's dashboard, look at usage and charges from October 17, 2026 onward. According to a news report, OpenCode's email gave 00:30 UTC on October 17 as the earliest possible exposure. Look for spikes, unfamiliar models and requests at hours you were not working. A spending cap set inside OpenCode does not limit requests made directly to the provider with a stolen key.
For what happens when a provider detects abuse on a stolen key, see stolen API keys and account bans.
Set limits at the provider before using the new key
Where the provider offers them, set a monthly spending limit and a usage alert on the account or project, and give the new key only the permissions and projects it needs. If the new key leaks, the damage stops at the limit.
If you were charged, keep records and contact both companies
According to news reports, OpenCode said refunds for unauthorised charges were underway, and some providers blocked keys or refunded credits used in the attack. Keep screenshots or exports of the usage records, then contact OpenCode and the provider through their official support pages. Share only masked key values.
What happened (from OpenCode's post and news reports)
OpenCode's official X account (@opencode, linked from opencode.ai) posted the following on October 17, 2026. Details beyond that post come from news reports of the company's email and statements, and are marked as such.
Night of October 16–17, 2026
A security breach affecting OpenCode Console (company). According to a news report of the email, the earliest possible exposure was 00:30 UTC on October 17.Morning of October 17
OpenCode emails all affected users (company).October 17
OpenCode posts on X that 1.4% of Console users were impacted and that the incident is ongoing.October 19
News outlets report the email's contents and statements by the company, including that the open-source app and command-line tool were not affected.As of October 31
No further official statement found on the company's status page or in a web search.
Linked in OpenCode Console
Keys stored in a Console workspace. The affected users were a subset of these (company); the keys may have been exposed (as reported)
Used only in the local app or CLI
Not affected, because those keys are not sent to the company (co-founder, as reported)
The split depends on where the key was stored, not on which AI provider issued it.
- What happened
- A security breach related to a large-scale distillation attack (company)
- Who
- 1.4% of OpenCode Console users, a subset who had external providers linked. All were emailed; no email means not impacted (company)
- What may be exposed
- API keys for external AI providers stored in Console workspaces (company email and statements, as reported)
- Not affected
- The open-source app and command-line tool (co-founder, as reported)
- Response
- Working with the AI providers concerned to remediate (company). Refunds for unauthorised charges underway (as reported)
- Cause
- Unauthorised access to an older system, according to news reports. Not explained in OpenCode's post
What is known and what is not
The state as of October 31, 2026. "As reported" means the item comes from news reports of the company's email or statements, not from a published OpenCode notice. When there is a follow-up, the status column will be updated.
| Item | What has been stated | Status |
|---|---|---|
| Share of Console users impacted | 1.4% | Confirmed (company) |
| Who was impacted | A subset of users who had external providers linked | Affected (company) |
| Notice to affected users | Email to all of them on the morning of October 17 | Confirmed (company) |
| External AI provider keys stored in Console workspaces | May have been exposed | Possible (the company's email to users described the keys as exposed, per reports) |
| Open-source app and command-line tool | Not affected | Not included (co-founder, as reported) |
| Number of users or keys | Only the percentage has been given | Not stated |
| How the keys were stored | Not explained | Not stated |
| Entry route | Unauthorised access to an older system (as reported); details not released | Under investigation |
| Misuse of keys | Users reported unauthorised charges (as reported) | Under investigation |
| Refunds | Underway, according to news reports; conditions not published | Not stated |
How to read this: the company's own post is short
OpenCode's published statement is a single post. The detail about keys comes from reports of its email. Until the company publishes more, act on the safer assumption: if you linked keys in Console, treat them as exposed.
What an exposed AI provider key can lead to
In general, an exposed AI provider key can be misused as follows. This does not mean all of it happened here.
- Requests run on your account, so you get the bill. Users reported unauthorised charges in this case, according to news reports
- The provider may suspend your account or project when it detects abuse, which also stops your own work
- Large volumes of model output collected at your expense, which is what "distillation" refers to in general
- Access to anything else the key can reach, such as files or fine-tuned models stored in the same project
A past example: in the Zeabur security incident, the platform said the attacker was mainly looking for API keys it could use directly, especially keys for AI services. This case may not follow the same course.
For teams that use AI coding tools
These are general checks based on what was exposed. They do not assess OpenCode's handling.
- Give each tool its own key. A separate key for each tool or service means you can revoke one without stopping everything, and you can see in the provider's usage page which tool a spike came from.
- Set the spending limit at the provider, not only in the tool. A limit inside a third-party tool does not apply to someone using the key directly.
- Keep a list of where each key is stored. When an incident like this happens, the first question is "which services hold a copy of this key?". A short list makes revoking fast (see: what's actually dangerous about .env and API keys).
Sources (public records)
The facts in this article are based on the public information below. The entry route, which the company has not explained, is not speculated about.
- OpenCode, official X account (@opencode, linked from opencode.ai), post of October 17, 2026 — x.com
- TechCrunch, "Hackers hit AI coding tool OpenCode, running up charges on users' AI accounts" (October 19, 2026). Source for the older system, the co-founder's statement, the advice to revoke keys and refunds — techcrunch.com
- BleepingComputer, "OpenCode hack exposes API keys amid AI distillation attacks" (October 19, 2026). Source for the earliest possible exposure time in the email and users' reports of charges — bleepingcomputer.com
Update history
2026-10-31: First version. Based on OpenCode's October 17 post on X and reports by TechCrunch and BleepingComputer on October 19. As of October 31, no further official statement was found on the company's status page or in a web search. This article will be updated when there is a follow-up.
Read next
- Same type of incident: Zeabur security incident (October 2026) / the LiteLLM 1.107.0 compromise
- Why AI keys are targeted: defending against AI-driven cyberattacks
- After a key is stolen: stolen API keys and account bans
- Other 2026 incidents: Data breaches and cyberattacks of 2026