Unknown attackers gained access to the API keys of users of the OpenCode service, leading to unplanned charges on third-party platforms, the website infohub.kz reports.
The attack targeted the legacy OpenCode Console system, which is used to manage software development processes with the help of artificial intelligence. About 1.4% of the platform's users were affected in the incident. The attackers stole stored API keys that were used to connect to AI services from OpenAI, Anthropic and Google.
Developers were hit with unexpected bills: on social media platforms X and Reddit, users began reporting unauthorized charges en masse — ranging from hundreds to more than $1,000.
According to OpenCode co-founder Dax Raad, the company has linked the incident to a large-scale AI model distillation attack. During the investigation, the team encountered an unusual obstacle: the AI systems' protective algorithms refused to analyze the malicious code, treating the engineers' requests as dangerous. The problem was resolved only after obtaining special access to OpenAI's Daybreak Blue program with relaxed safety filters.
Affected developers are now having their compromised access keys revoked and are being issued refunds for the charges. The local OpenCode client itself was not affected in the breach.