Play store
A cyberattack targeting an older system of OpenCode, a popular AI coding tool, has affected 1.4 percent of its users, with some reporting stolen API keys and unauthorised use of AI services that resulted in bills exceeding $1,000.
According to technology news outlet TechCrunch, OpenCode disclosed the incident last Saturday. The company suspects the attack may be linked to a large-scale attempt to distil AI models, a process in which one model’s outputs are used to train another. However, it remains unclear whether model distillation was the motive or whether the incident was a conventional security breach.
OpenCode co-founder Dax Raad said the attack affected only the company’s console, leaving its open-source application and command-line tool untouched. Saved API keys in the console may have been exposed, potentially allowing attackers to access third-party AI services without authorisation.
The affected services included those provided by OpenAI, Anthropic and Google.
Following the incident, Anthropic and Google disabled API keys suspected of being compromised. OpenAI worked with OpenCode to restore services and began refunding credits consumed during the attack. OpenCode has also started reimbursing users for unauthorised charges.
Raad said the attackers repeatedly attempted to breach the system over a 48-hour period. The company used AI agents to help analyse the security incident, but several AI models refused to examine malicious code because of their safety restrictions.
OpenCode subsequently gained access to models suited to security research through OpenAI’s Daybreak Blue programme.
The incident highlights a challenge in using AI to defend against cyberattacks: safety guardrails designed to prevent harmful use can also obstruct legitimate security research, which often requires examining malicious code. AI platform Hugging Face has previously reported a similar problem.
OpenCode has not disclosed how the attackers penetrated the legacy system, how many API keys were exposed or how many were used fraudulently. The investigation is ongoing.
Affected users have been advised to visit the relevant AI service providers’ platforms, revoke potentially compromised API keys and generate new ones to prevent further unauthorised access.