Account published describing 48-hour sustained attack, defender use of AI agents, safety guardrail failures, and access to Daybreak Blue defensive program.
Following·Day 7·5 sources
OpenCode Breach Exposes AI API Keys in Distillation Attack
The gist
Developers storing API keys in hosted consoles face direct financial exposure when those platforms are breached, with costs billed to them while attackers harvest model outputs. The incident shows that safety guardrails in frontier models can block defenders during active incidents, a problem that also affected Hugging Face.
The full picture
OpenCode disclosed a security breach affecting 1.4% of its Console users who had external AI providers configured, with API keys for OpenAI, Anthropic, and Google potentially exposed. The breach was attributed to a large-scale distillation attack. OpenCode notified affected users by email and faced sustained attack attempts for at least 48 hours after the initial incident. CellCog's analysis explains the attackers' approach: a real customer's API key is verified, has payment history, and produces traffic that blends in as ordinary, so distillation traffic is charged to the victim rather than the attacker. OpenCode has not disclosed which models were queried, how many keys were affected in total, or the total fraudulent spend, and its notice promised follow-up updates.
How it developed
OpenCode breach attributed to distillation attack reported; roughly 1.4% of Console users potentially affected with API keys for OpenAI, Anthropic, and Google exposed
OpenCode disclosed the breach, notified affected users by email, and attributed it to a large-scale distillation attack affecting 1.4% of Console users.
Sources
Rohan Paul Twitter
Zvi's AI Roundups
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free