Anomaly OpenCode Console · Breach

OpenCode Console breach: OpenAI, Anthropic and Google revoke exposed keys

Oct 16, 2026 · 1 min read

OpenCode Console, Anomaly's hosted service, suffered a breach disclosed on October 17, 2026, linked by OpenCode itself to a large-scale distillation attack. OpenAI, Anthropic and Google keys saved in about 1.4% of workspaces may have been exposed starting October 16, 8:30 PM ET; all three providers revoked or blocked affected keys, OpenAI refunded drained credits and OpenCode is paying refunds. Rotate every key linked to Console, revoke OAuth grants and review usage and billing since October 16, 8:30 PM ET; the incident has not been declared resolved.

By HIFY Radar newsroom Updated 10/26/2026 01:25 PM Get this every day →← Back to the feed 3 sources

HighBreachNo software to update

In this article

What happened
Why it matters
What to do
For the technical team
Timeline
Sources

What happened

OpenCode (Anomaly) announced on October 17, 2026 a breach of OpenCode Console, its hosted service, which it links to a large-scale AI distillation attack. OpenAI, Anthropic and Google keys saved in workspaces may have been exposed since October 16, 8:30 PM ET. About 1.4% of Console users were affected and emailed; OpenCode says users who got no email were not impacted.

Why it matters

Exposed keys allow fraudulent use of AI services and unauthorized charges. OpenAI says it refunded drained credits; Anthropic was still restoring drained credits the week of October 20; OpenCode says refunds are underway. OpenCode calls the incident ongoing and reports hundreds of attack attempts per day, including OAuth phishing.

What to do

Rotate now every provider key linked to Console: create a new key and delete the old one at each provider. Revoke provider connections in the workspace and any OAuth and app grants, and review usage and billing since October 16, 8:30 PM ET. Set per-key spend limits and turn on alerts; anyone drained who missed the email should contact OpenCode.

For the technical team

Affected: Anomaly OpenCode Console (serviço hospedado) · OpenCode Console (hosted service by Anomaly): workspaces with saved provider keys (OpenAI, Anthropic, Google; rotation guidance also names OpenRouter and xAI). About 1.4% of Console users, emailed the morning of October 17. No specific versions cited in the sources.
Exploitation: Exposure confirmed per key: OpenCode's follow-up email lists each exposed key by workspace and last 4 characters, with earliest possible exposure October 16, 2026, 8:30 PM ET (00:30 UTC Oct 17). Anthropic (Lydia Hallie) revoked the Claude keys it believes were affected; Google (Evan Otero) blocked exposed Gemini keys; OpenAI (Tibo) restored affected customers and refunded credits, and the exposed OpenAI keys were revoked together with OpenCode. Dax reports 48 hours of continuous attempts (Oct 18), hundreds of different attempts per day (Oct 21) and other similar products hit.
Fix: No product fix announced. Mitigations cited: create a new key at each provider and delete the old one (a key suddenly returning 401 was likely revoked by the provider); revoke provider connections in the workspace plus OAuth and app grants; re-link only with a fresh, OpenCode-only, narrowly scoped key; hard monthly and per-key spend limits; usage and billing alerts. OpenCode says it is hardening its systems and gained OpenAI's "Daybreak Blue" defensive-security access.
Vector: Breach of the hosted OpenCode Console service; the local OpenCode app was not affected and keys used with it locally "never come to us", according to OpenCode's Dax. OpenCode ties the case to a "large scale distillation attack"; the initial entry vector is not described in the sources. Post-breach phishing included a real PlanetScale OAuth link rigged to send credentials to the attacker and an attempt via the enterprise contact form aimed at a locally running OpenCode server.
More details: Timeline: OpenCode's initial post Oct 17, 1:56 PM ET (edited 2:08 PM); Anthropic (Lydia Hallie) at 5:04 PM; Google (Evan Otero) at 6:53 PM, clarifying Oct 19 it was the Console; Dax (@thdxr) clarified Console vs app at 5:38 PM. On Oct 18, Dax cited 48h of continuous attempts, frontier models refusing to help mid-investigation and OpenAI Daybreak Blue access. On Oct 19, he said other similar products were also hit and OpenCode is talking to AI labs about better protocols. On Oct 20, Anthropic was restoring drained credits; OpenCode will send them itself if Anthropic has not finished. The follow-up email sets earliest exposure at Oct 16, 8:30 PM ET and says no other data appears to have been exposed. OpenAI accounts banned during cleanup should be reinstated. A cited workspace suspension was routine anti-fraud/anti-distillation enforcement. Missed refunds: contact Dax at the email he posted. ctxsec tracking ended Oct 25, 8:15 AM ET, with no resolution declared by OpenCode.