The OpenCode tool suffered an intrusion that exposed some users’ API keys, allowing hackers to use their AI accounts at their expense. The incident affects 1.4% of users of the OpenCode Console, the project’s online service, and several developers report unauthorized expenses that could exceed $1,000.
Compromised API keys
In a messageOpenCode explained that a third party had accessed an old system that the team was in the process of decommissioning. External vendor API keys stored in some OpenCode Console workspaces have been exposed. These identifiers allow an application to query AI models and charge the key owner for the queries. OpenCode cites OpenAI, Anthropic and Google among the vendors involved in the operations.
According to Dax Raad, co-founder of the project, the compromise affects the OpenCode Console and not the open source application itself or its command line tool. The keys in question are those stored in the hosted service’s workspaces. OpenCode indicates that it has notified by email all those identified as affected, that is, 1.4% of Console users.
After the intrusion, users reported unusual consumption alerts in their OpenAI, Anthropic or Google accounts. Some cite several hundred dollars in expenses, while others report amounts exceeding $1,000. These fees can add up quickly when an API key is used to send a large number of requests to expensive models. OpenCode said refunds related to unauthorized use were being made, while OpenAI worked with the project to restore access and refund credits consumed during the attack.
OpenCode describes the incident as“large scale distillation attack”. Distillation involves exploiting the responses of a successful model to train or improve a less expensive model. The stolen keys could be used to generate these responses in large quantities. However, the technical details of the intrusion have not been made public.
What should affected OpenCode users do?
OpenCode asks those notified by email to revoke their compromised keys directly from each provider and then generate new ones. It is also a good idea to check billing dashboards and usage logs to detect unknown requests, consumption spikes, or suspended projects. Keys must be renewed at the providers themselves: simply changing an OpenCode password is not enough to invalidate an exposed API key.
The team says the incident is still ongoing and has not yet explained how the hackers penetrated the old system, how long they had access to it, or what precise data they accessed. This lack of technical explanations currently prevents us from establishing the full extent of the hack. However, the case reminds us of the risk associated with the centralized storage of API keys that give access to various AI services: their exposure can lead to access leaks and significant costs for users.
