OpenCode: AI incident

In October 2026, unknown individuals gained access to a legacy system of the OpenCode Console, which was already in the process of being shut down at that time. This allowed them to access API keys belonging to users who had linked external AI providers. According to OpenCode, 1.4 percent of Console users were affected. Keys for services such as OpenAI, Anthropic, and Google were reportedly exposed.

OpenCode confirmed the incident on October 17, 2026, and described it in a statement as “a large-scale distillation attack that impacted 1.4% of OpenCode Console users.” The company did not explain exactly how the attackers carried out the attack or how the keys were stored. According to the company, anyone who did not receive a notification from OpenCode is not affected.

Several users reported being billed for usage they did not initiate. OpenCode has not yet confirmed whether the exposed keys were actually misused. The company stated that it is working with Google, OpenAI, and Anthropic on countermeasures. There is no public identification of the attackers.

The security firm Cyera documented the case on October 17, 2026 (Authors: Adam Ochayon and Yuval Yogev) and is maintaining it as a continuously updated report. It references an Anthropic report from September 2026, without claiming a connection to the OpenCode incident. This report describes a campaign by Chinese AI labs that intercepted approximately 190 million Claude exchanges, in part via illegally obtained API keys from real companies and individuals.

Cyera advises affected organizations to revoke and reissue keys not only in the OpenCode Console but directly in each provider’s dashboard. Users should access the dashboards directly, rather than via links in notification emails, as such messages often serve as phishing lures.

The company also advises resetting OpenCode passwords, regenerating Console API keys, and reviewing the providers’ usage and billing data since October 16, 2026, for outliers, unknown models, or activity at unusual times. Spending limits in the Console only protect against abuse via OpenCode itself, not against a key used directly against the provider.