The drama in startup land exposes a legal question about AI agents after Primcloud founder Josh Manders accused Railway employee Jake Runzer of trying to breach his company’s infrastructure. The accusation appeared on November 3, 2026, and Runzer said an automated testing effort had gone too far.
Runzer, whom Railway CEO Jake Cooper tagged in a response, acknowledged the activity. He said he had been running agents to create signups and test similar platforms, then halted the harness he was designing because it was reaching too far. Cooper apologized and said Railway would ban testing models on external organizations unless it was performed manually.
Startup land exposes questions about intent
The episode raises the question of who could be legally responsible when an AI agent attempts to access a computer system. Mark Rasch, general counsel of threat advisory firm Unit 221B, said prosecutors would need to prove that an agent attempted illegal access and that its user intended that outcome.
Proving intent could be difficult in security testing, Rasch said. In a bug bounty program, companies offer security researchers a deal to find system flaws, while the researchers do not intend to break the law. But Rasch said an agent could find a possible way in and decide to try it.
That possibility also raises negligence questions. Elimu Kajunju, a cybersecurity lawyer at Rimon Law, said defendants might try to shift blame to the companies that built their AI agents. Kajunju said such cases were becoming harder to defend because similar incidents had happened multiple times.
Rasch and Kajunju said legal precedent was still forming. Their comments point to an unresolved issue for companies and people using agents: how the law assigns responsibility when automated tools act beyond their operators’ intentions.
Railway says employee halted the tool
Cooper later described the incident as involving a tool an employee created on personal time for proactive identification and reporting of bug bounties in the broader community. He said it used an externally hosted service running on DeepSeek, was used once, and was halted proactively within 15 minutes of starting.
Railway said it did not condone the actions, had addressed the incident with the employee, and was reinforcing expectations for employee conduct. Cooper also said the company applies strict guidelines to internal red teaming and follows responsible disclosure requirements for bug bounty reports.
Manders disputed the bug bounty framing. He said Primcloud did not have a bug bounty program and that its acceptable use policy treats attempts to probe, scan, or test its vulnerabilities as security violations. The disagreement leaves a central fact in the dispute clear: the two companies described the activity under different terms.
It remains unclear whether Manders will sue. He, Runzer, and Cooper did not respond to Business Insider’s requests for comment. Cooper wrote that operators are responsible for what their machinery does.
The dispute adds to coverage of alleged misconduct involving technology companies, including the AppliedAI founder’s guilty plea and the 777 Partners founder’s fraud case. Questions about autonomous tools also connect to a federal oversight probe into an AI company plan. Business Insider’s report on the Railway and Primcloud dispute describes an incident with no announced lawsuit and legal standards that Rasch and Kajunju said are still developing.
