OpenCode Console Breach Exposes API Keys, Driving Unauthorized AI Spend

OpenCode disclosed on Saturday that a large‑scale distillation attack breached an older component of its online console, exposing the API keys that some developers store in their workspaces. The incident affected roughly 1.4% of console users, while the open‑source command‑line tool remained untouched.

Developers quickly reported spending alerts from OpenAI, Anthropic, and Google Cloud, with individual accounts seeing charges ranging from a few hundred to over one thousand dollars. In response, Anthropic and Google revoked the compromised keys, and OpenAI collaborated with OpenCode to restore access and issue refunds for the abused credits.

OpenCode co‑founder Dax Raad noted that the team faced continuous intrusion attempts for 48 hours, forcing them to run defensive AI agents around the clock. Those agents initially helped analyze the malicious code but began refusing requests once safety filters were triggered, slowing the investigation.

Why it matters for GPU and AI infrastructure

The breach highlights a structural asymmetry: security teams need unrestricted model access to dissect attacks, yet commercial guardrails often block legitimate defensive work. Vetted programs such as OpenAI’s Daybreak Blue, which grant lower‑refusal models to approved responders, become critical for rapid incident response on GPU‑heavy platforms.

As AI‑driven coding assistants proliferate, protecting credential storage and ensuring privileged, safe model access will be essential for any infrastructure provider — including aigpu — that hosts or orchestrates large‑scale AI workloads.