OpenCode has revealed that around 1.4% of its Console users are affected by a security breach that may have exposed third-party API keys stored in their workspaces. The exposed credentials are said to include service keys for providers like OpenAI and Anthropic. This issue stems from unauthorized access to a legacy system the company was already phasing out. Users who did not receive a notice are not believed to be impacted.
The company issued a public advisory on October 17 describing the incident. It says the breach is tied to what it’s calling a “large-scale distillation” attack, and that it is collaborating with OpenAI, Anthropic, and Google to address any downstream risks. It has urged affected customers to revoke and replace compromised keys, update applications relying on them, and review usage logs to detect unusual activity.
Nature of Exposure and Gaps in Disclosure
The breach allegedly involves a legacy backend system previously scheduled for decommissioning. The exposed data includes stored keys for third-party providers in certain user workspaces. Although the advisory warns of possible exposure, it stops short of confirming widespread misuse or theft of keys. OpenCode has neither quantified exactly how many keys were affected nor disclosed whether attackers managed to exploit them.
While the term “distillation” is used to describe the attack vector—where attackers train AI models by harvesting responses from another model—the disclosure is vague on whether such model requests were made with exposed credentials. There is no detailed timeline of the attack’s discovery, no named threat actor, and no confirmed numbers for loss or usage so far.
What Users Should Do Now
Affected users are instructed to revoke and replace any exposed API keys via their provider dashboards, update any software that depends on those keys, and audit recent key usage for anomalies. Reviewing billing statements and reviewing provider dashboards will help detect any signs of unauthorized access.
OpenCode says more details will follow when it prepares a full technical report. For now, however, it is not sharing how attackers gained access, how long the breach remained undetected, or whether the attackers exploited model-serving endpoints using compromised credentials.
Implications and What to Watch
This breach is an important illustration of the risks posed by maintaining legacy systems, especially ones housing sensitive credentials. As many AI tools increasingly rely on third-party integrations, exposure of API keys can lead to downstream data misuse and unauthorized access. The distillation angle raises additional concerns about model theft and output leakage.
Going forward, users should watch for the full incident response from OpenCode, including technical attribution and scope of exposure. Providers like OpenAI, Anthropic, and Google also will be under pressure to clarify how exposed keys might have been used—or misused—in model training, data extraction, or other backend operations.
Ultimately, this case reinforces the need for robust key management practices: minimize storage, enforce rotation, use least-privilege access, monitor activity, and phase out legacy services with sensitive assets. The integrity of AI models and services depends not just on new innovations but on disciplined operational hygiene too.
