Attackers broke into an older OpenCode system at the weekend and used stored API keys to run up charges on users’ AI accounts, the open-source coding tool’s team says. TechCrunch reported it on 19 October.
The system was part of OpenCode Console, where some users keep keys for AI services. OpenCode says 1.4 percent of Console users were affected and has emailed them. Anyone without an email was not.
Developers reported spending alerts and unauthorised charges on OpenAI, Anthropic and Google accounts. Some were in the hundreds of dollars, and a few were above $1,000.
Anthropic and Google blocked the keys they believed were exposed. OpenAI worked with OpenCode to restore access and refund credits. OpenCode says refunds are under way, and told affected users to revoke and replace their keys with each provider. Co-founder Dax Raad says the open-source app and command-line tool were not affected.
OpenCode called the incident a large-scale distillation attack, meaning mass copying of a model’s outputs to train another, but did not explain how that worked. Some sceptics say the label is unproven. Leaked keys could help users in countries with restricted access to US models generate the volumes distillation needs.
Raad said attackers had been trying to get in nonstop for 48 hours, so the team ran AI defensive agents around the clock. Commercial models helped at first, then refused once the malicious code tripped their safety filters. Access to OpenAI’s Daybreak Blue programme, which gives vetted users models less likely to refuse defensive work, then improved the investigation.
Still unanswered: how the attackers got into the older system, how the keys were stored, and how far the misuse went. OpenCode says the incident is ongoing.
