OpenCode reports a Console breach affecting 1.4% of users

• OpenCode reported a breach affecting 1.4% of Console users; the company says everyone affected was emailed.
• External-provider keys in a legacy hosted system may have been exposed, according to the customer notification.
• The reported scope is Console: a maintainer said keys used directly in OpenCode “never come to us.”
• Refunds were already underway, according to the maintainer’s later reply.

Console’s provider documentation describes a shared gateway where workspace members can use models without ever seeing the underlying provider keys. The earliest possible exposure was October 17, 2026, at 00:30 UTC, according to one affected user’s follow-up notification.

OpenCode attributed the exposure to unauthorized access to a legacy system it was decommissioning in its initial customer notification. A follow-up itemized affected keys by workspace and said no other data appeared exposed at that point.

Affected accounts were a subset of Console users with linked external providers, OpenCode said in its announcement. The company said anyone who had not received an email was unaffected and that it was working with Anthropic, OpenAI, and Google to remediate the incident.

OpenCode maintainer thdxr drew the product boundary explicitly:

Provider API keys added directly through the client’s /connect command are stored in ~/.local/share/opencode/auth.json, according to the client provider documentation. For login-based connections, thdxr said OpenCode defaults to OAuth tokens rather than long-lived API keys in another reply.

The shared branding did the disclosure no favors: people in the Hacker News discussion repeatedly asked whether the locally installed coding agent had been compromised.

OpenCode described the breach as related to a “large scale distillation attack” in its announcement, while saying the incident was ongoing and further details could not yet be shared.

The connection between stolen API keys, high token volume, and model training drew questions from natolambert. He asked the company whether the illegally obtained tokens were being used to train downstream models or for some other purpose:

Users reported unauthorized provider charges, including one report of $400 spent on an Anthropic account, according to kimmonismus. Reimbursement was being handled through direct contact with the maintainer:

Refunds were “already underway” and everyone affected had been contacted, thdxr said in a later reply. He described the additional outreach as an effort to catch people who might have missed an email.

OpenCode’s notification email asked affected customers to:

• Revoke and replace the keys through the relevant provider’s dashboard.
• Update applications using those keys.
• Review recent usage for unexpected activity.