Hackers attacked the open source of AI coding tool OpenCode and imposed charges on users' AI accounts without permission, TechCrunch reported on Oct. 19.

The attackers broke into an older OpenCode system, the report said. They compromised 1.4 percent of OpenCode Console user accounts. Some users had stored API keys there, which act like passwords for accessing AI services.

OpenCode said the incident is linked to a large-scale distillation attack. It did not disclose the specific method. Some skeptics say it exaggerated a simple breach.

On Reddit and X, developers have posted a string of cases of unauthorized payments. TechCrunch reported that accounts were charged hundreds of dollars and some cases exceeded $1,000.

Anthropic and Google blocked API keys that appeared to have been exposed. OpenCode is also processing refunds for unauthorized charges.

Dax Raad (닥스 라드), OpenCode's co-founder, said the leak was limited to the console and that open source apps and command-line tools were not affected. He said the attackers kept trying to break in without a break for 48 hours.

The OpenCode team ran a defensive AI agent around the clock, and AI models that were helping the analysis refused to continue midway after judging the malicious code to be a dangerous request.

OpenCode later used OpenAI's "Daybreak Blue" program to strengthen its investigative capabilities. Through the Daybreak Blue program, OpenAI provides verified users with AI models that are relatively less likely to refuse defensive security requests. In July, Hugging Face also had commercial AI models analyze intrusion logs but was refused.

OpenCode has not yet disclosed how the attackers gained access to the older system. It told affected users to revoke their keys with each provider and have new ones issued. The company said the incident is still ongoing.