OpenCode Console breach exposes AI API keys and costs
A breach affecting 1.4% of OpenCode Console users triggered unauthorized AI charges and exposed a difficult gap in defensive model access.
What happened
OpenCode said attackers compromised an older system connected to OpenCode Console, its hosted service for developers who store provider credentials and run coding workflows. The company said 1.4% of Console users were affected and that the open-source OpenCode application and command-line tool were not compromised.
Users reported sudden spending alerts and unauthorized charges across OpenAI, Anthropic and Google accounts. Some losses reached hundreds of dollars, with a few reports exceeding $1,000. Anthropic and Google blocked exposed keys, while OpenAI worked with OpenCode on access restoration and refunds. OpenCode advised affected customers to revoke and replace credentials directly with each provider.
The incident became more complicated because OpenCode initially described it as a large-scale model-distillation attack, while outside observers said the evidence publicly available so far established a credential breach but not the attackers’ ultimate purpose. OpenCode also said the assault continued for roughly 48 hours, forcing its team to keep defensive agents operating around the clock.
Why it matters
This is a security incident, but its significance extends beyond one developer platform. AI coding tools increasingly sit between users and several model providers, turning stored API keys into a concentrated target. A compromise can therefore create financial exposure across multiple vendors at once, even when the underlying coding tool remains intact.
The episode also exposed a practical safety dilemma: models may refuse to inspect malicious payloads even when an incident-response team is investigating them. OpenCode later obtained access to OpenAI’s Daybreak Blue program, whose vetted models are designed to be more useful for defensive security work. That distinction could become an important operational requirement for companies deploying agentic coding systems.
Uncle Cat take: OpenCode’s 1.4% figure understates the systemic risk: one console breach can turn three model providers into a single attacker-controlled billing surface.