Hackers breach OpenCode platform, triggering unauthorized AI service charges for users

OPA WIRE, Oct 19 (OPA) — OpenCode, an open-source AI coding agent, confirmed a security breach occurred overnight Saturday, affecting a small fraction of its online platform users. The company stated that attackers gained access to an older system where some users stored API keys used to interface with third-party AI services. Affected individuals reported receiving automated spending alerts, with some unauthorized charges exceeding $1,000.

Co-founder Dax Raad confirmed the breach was limited to the OpenCode Console, noting that the core open-source application and command-line tools remained secure. Anthropic and Google have since blocked the compromised keys, while OpenAI is coordinating with OpenCode to restore access and facilitate refunds. The company has instructed all impacted users to revoke their existing keys and generate new credentials directly through their respective AI providers.

The incident highlighted operational friction for security teams relying on AI-driven defense tools. During the 48-hour attack window, OpenCode developers reported that their defensive AI models frequently refused to analyze malicious code, as the payloads triggered built-in safety filters. The team eventually gained access to OpenAI’s Daybreak Blue program, which provides vetted users with models that have fewer restrictions, allowing for more effective incident investigation.

This breach follows a series of AI-related security incidents, including reports of autonomous agents probing systems for vulnerabilities. David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, noted that current safety guardrails can inadvertently impede incident responders. OpenCode continues to investigate the breach, though the company has not yet disclosed the specific method used to gain initial access to its legacy systems.