OpenCode API key breach: 1.4% of Console Users, $1,700 Bills

The OpenCode API key breach is the AI-tools security story of the week. The open-source coding agent told about 1.4% of its Console users that provider keys they had linked for OpenAI, Anthropic or Google may have been exposed, and some of them woke up to hundreds or even more than $1,700 in charges they did not make. OpenCode blames a large-scale distillation attack.

What actually happened

OpenCode is a model-agnostic coding agent that runs in a terminal, desktop app or editor. Its hosted Console lets teams store keys for outside model providers so everyone can use them in one place. On Saturday, the project posted that it had suffered "a security breach last night related to a large scale distillation attack."

The Hacker News reported that the intrusion hit a legacy system that was in the process of being shut down, on the night of October 16. Affected users got an email titled "Action Required: Potential provider key breach." OpenCode said it was working with Anthropic, OpenAI and Google, and the notice to users mentioned the possibility of reimbursement.

TechCrunch and devops.com also covered the incident on Monday, and users kept posting screenshots of surprise invoices.

Why "distillation" matters here

Distillation is a normal machine-learning technique: you train a smaller model on answers from a bigger one. It becomes an attack when someone does it at scale on stolen access, so the bill lands on the victim and the frontier lab's terms are broken at the same time. Stolen keys are the perfect tool, because the traffic looks like a paying customer.

This fits a pattern. Labs have been publicly accusing rivals of copying model outputs, as in our piece on the OpenAI Moonshot distillation claims. The difference here is that ordinary developers paid for it.

The damage so far

Charges reported by users ranged from a few hundred dollars to more than $1,700 overnight. One developer posted about $281 on OpenAI and $168 on Anthropic; another reported around $1,700. Some users said Google suspended their cloud projects after detecting "abusive activity consistent with hijacked resources," which is its own headache.

Then there is the twist. Raad said OpenCode's own investigation stalled when the AI models it was using refused requests that looked like hacking, and things stabilized only after the team got approval to use OpenAI's Daybreak cybersecurity model on Saturday. Safety filters blocking the defenders is a very 2026 problem, and the same theme came up in our OpenAI $300 bug bounty story about who gets to test what.

Your AI coding agent did not get hacked. Your API key did, and your card was attached.

The part people miss

According to The Hacker News, deleting a key inside the OpenCode Console does not revoke it at the provider. The key keeps working until you kill it on the OpenAI, Anthropic or Google side. That is the single most important fix in the whole OpenCode API key breach, and it is easy to get wrong if you only clean up the app.

What to do as a reader

If you ever linked a provider key in OpenCode Console, assume it is burned. Revoke and rotate it directly in your OpenAI, Anthropic and Google dashboards, set hard spending limits, and check usage logs back to October 16. If you see charges you did not make, open a billing dispute with the provider and keep the OpenCode email as evidence.

Going forward, give every tool its own key with a low monthly cap, so one breach cannot drain your whole account. Keys used only in the local CLI or desktop app were not part of this incident, but the same rule applies there. The OpenCode API key breach is a reminder that convenience and credential sprawl go together.

Sources: BleepingComputer report and The Hacker News report.