Hackers Breach OpenCode AI Coding Agent, Drain Users' API Credits

On the night of October 19, hackers broke into OpenCode Console, the web dashboard for OpenCode. OpenCode is an open source AI coding agent used as an alternative to Claude Code and Cursor, and users store their API keys there for access to paid AI services. The breach affected 1.4% of Console users. The OpenCode app itself and its command-line tool were not affected. The stolen keys let the attackers run up charges on victims' OpenAI, Anthropic, and Google AI accounts, from hundreds of dollars to more than a thousand per account.

The most awkward part: the OpenCode team itself struggled at first to make sense of the attack. Commercial AI models refused to analyze the malicious code, because it triggered their own safety filters. Only after the developers got access to a less restricted version through an OpenAI program could they figure out what happened. After an incident like this, issuing an agent a key scoped to a single task, rather than one for the whole account, is no longer just a precaution. It is a necessity.