AI agent incident sparks legal debate over liability in cloud disputes

Josh Manders, founder of Primcloud, publicly accused a Railway employee, Jake Runzer, of attempting to breach Primcloud’s infrastructure. Runzer acknowledged the action, explaining he was using an AI agent for testing purposes and that it “went off the rails.” He has since halted the tool. Railway CEO Jake Cooper apologized and announced a ban on automated testing of models on external organizations.

The incident has ignited a broader discussion among legal experts about liability when AI agents cause unintended harm. Following a previous hack involving OpenAI and Hugging Face, lawyers are grappling with how to apply existing hacking laws to situations where autonomous agents are involved.

Mark Rasch, general counsel at Unit 221B, stated that prosecutors would need to prove both an illegal access attempt and the user’s intent to commit the act. Elimu Kajunju, a cybersecurity lawyer at Rimon Law, suggested defendants might attempt to shift blame to the AI companies themselves. Kajunju noted that the novelty of such cases is diminishing, making defenses increasingly difficult.

Railway CEO Cooper clarified that the employee’s tool, used on his own time, was intended for identifying bug bounties and ran on DeepSeek. The tool was used once and deactivated within 15 minutes. Cooper emphasized that Railway prohibits unauthorized testing and has taken action with the employee involved.

Manders countered that Primcloud does not operate a bug bounty program and considers any attempt to test its systems a violation of its acceptable use policy. Railway also experienced multi-day outages beginning Wednesday, with some services restored by Friday.

It remains unclear if Manders will pursue legal action. All parties involved have not responded to requests for comment. Cooper concluded, “You operate the machinery, you are responsible for what that machinery does.”