OpenCode says attackers broke into an older system, exposing API keys belonging to 1.4% of users of OpenCode Console, its online platform. Some users reported unauthorized charges on their OpenAI, Anthropic and Google AI accounts, including bills in the hundreds of dollars and, in a few cases, above $1,000.
The company said it emailed everyone affected and told them to revoke and replace their keys directly with each provider. Anthropic and Google blocked keys they believed were exposed. OpenAI worked with OpenCode to restore access and refund customers for credits used in the attack; OpenCode says refunds are underway.
The breach was limited to Console
OpenCode co-founder Dax Raad said the open-source app and command-line tool were not affected. The exposed keys were in Console, where some users had saved them in workspaces. OpenCode has not explained how attackers entered the older system, how the keys were stored or the full extent of their use. The company says the incident is ongoing.
OpenCode described the incident as a large-scale “distillation attack,” a process that uses a model’s outputs to help train another model. But it has not explained how that process was carried out, and some observers questioned whether the breach’s connection to distillation has been established.
Safety filters complicated the response
Raad said attackers kept trying to break into OpenCode’s systems and that models the team used to examine suspicious code began refusing requests when safety filters were triggered. He said access to OpenAI’s Daybreak Blue program, which offers vetted users models less likely to refuse defensive security work, improved the investigation.
The episode echoes a problem Hugging Face described after an intrusion in July: safety systems may block defenders from analyzing the same malicious code that attackers use. That does not establish that the filters caused this breach, but it shows how model restrictions can complicate incident response. OpenCode has not disclosed how the attackers got in or how much unauthorized use occurred.
Why it matters
For developers who stored provider API keys in OpenCode Console, the immediate step is to revoke and replace those keys with each AI provider; OpenCode says refunds for unauthorized charges are underway. The company has not said how the attackers gained access or disclosed the full amount of misuse, so the overall exposure remains unclear.
