News · SI
OpenCode says a breach tied to a large-scale distillation attack hit 1.4% of its hosted Console users, the ones who had linked their own model provider accounts. Anthropic says it has revoked Claude API keys it believes were affected. If you saved a provider key in OpenCode Console, rotate it and check your bills.
OpenCode says a breach of its hosted Console service exposed model provider keys belonging to some of its users, and the company is working with Anthropic, OpenAI and Google to clean it up[1].
01 / 06
In a post on X on 17 October (UTC), OpenCode said it "had a security breach last night related to a large scale distillation attack that impacted 1.4% of OpenCode Console users"[1]. It says it emailed all of them that morning, that the affected group was a subset of users who had linked external providers, and that anyone who didn't get an email wasn't impacted. Because the incident is ongoing, it says it can't share more details yet[1].
The email to affected users, as reported by The Latent, said someone had gained "unauthorized access to a legacy system that we were in the process of decommissioning" and that provider keys saved in a workspace "may have been exposed"[5].
02 / 06
The breach is about OpenCode Console, the company's hosted service, where workspace admins can connect their own OpenAI, Anthropic or Google accounts and Console stores those keys on its servers[5]. OpenCode's Dax Raad stressed the distinction: "this is the OpenCode Console, not OpenCode itself, those keys never come to us"[3].
So if you only run the open-source agent locally with your own keys, OpenCode's account is that those keys weren't part of this.
03 / 06
Anthropic moved quickly. Lydia Hallie, who works on Claude Code at Anthropic, posted that Anthropic has been working with OpenCode on the breach "and have revoked Claude API keys we believe were affected", and suggested anyone who saved a key in OpenCode create a new one to be safe[2].
Before OpenCode said anything publicly, some users were already noticing charges they didn't make. The Latent collected reports of keys used only in OpenCode suddenly racking up hundreds of dollars on OpenAI and Anthropic, with one user putting his losses at US$1,700; it said it couldn't independently verify them[5].
The pressure hasn't stopped. A day later Raad wrote that attackers had been "non stop trying to attack our systems for the past 48 hours", adding that everything was getting hardened as a result[4].
04 / 06
If you use OpenCode Console with your own provider keys:
05 / 06
OpenCode hasn't said how many accounts 1.4% represents, how the attacker got in, how the breach connects to the distillation attack it describes, or whether it will reimburse anyone for stolen usage[5].
06 / 06
Coding agents increasingly sit between developers and several paid model providers at once, and hosted dashboards that hold those keys become a single place to steal many of them. Long-lived provider keys are also exactly what someone needs to pull large volumes of frontier-model output cheaply. The practical lesson is the boring one: give each tool its own key, cap its spend, and rotate it the moment a vendor says anything about a breach.
Sources
FAQ
OpenCode, the open-source AI coding agent, said on X on 17 October 2026 (UTC) that it had a security breach the night before, related to a large-scale distillation attack, affecting 1.4% of OpenCode Console users. It says the affected users were a subset who had linked external model providers, and that it emailed all of them.
OpenCode says no. Dax Raad of OpenCode wrote that this is the OpenCode Console, the company's hosted service, and not OpenCode itself, adding that keys used in the open-source tool 'never come to us'. The disclosure concerns provider keys saved in Console workspaces.
OpenCode says it emailed every affected user and that if you did not get an email, you were not impacted. The email, as reported by The Latent, had the subject 'Action Required: Potential provider key breach'.
If you got the email, revoke and replace every provider key you linked to Console in each provider's own dashboard, update anything that uses the old keys, and review usage and billing for activity you don't recognise. Anthropic says it has already revoked Claude API keys it believes were affected and suggests creating a new key if you saved one in OpenCode.
OpenCode hasn't explained the link. In general, distillation means training one model on the outputs of another, which takes a lot of paid model calls. Some users reported unexpected charges on keys they only used with OpenCode, from a few hundred dollars up to about US$1,700, which The Latent said it could not independently verify.
OpenCode hasn't said. It also hasn't disclosed how many accounts 1.4% represents, how the attacker got in, or the total cost of the unauthorised usage. It says the incident is ongoing.
That’s the story
Spectrum Web Co ships SI and Web3 products, places talent and builds social narratives as monthly services.