Autonomous AI agents are moving from helpful assistants to independent decision-makers inside large organisations, and the question of who answers for their mistakes is becoming urgent. In a new analysis, Norm AI argues that companies deploying these systems can no longer treat accountability as an afterthought.
According to Norm AI, AI is no longer limited to co-pilots, chatbots or on-demand content generation. Businesses are already putting agents into high-stakes roles with little or no human involvement, and that trend is set to accelerate as the technology improves. Inside large firms, knowledge and authority are spread across management layers, employees, contractors and software systems. As agents take on more of that work, people are shifting from doing tasks to overseeing machines. Norm AI notes that the ability to explain, reproduce or halt what an agent has done is becoming both harder and more important.
The challenge deepens with multi-agent systems. Norm AI highlights that one agent’s output becomes another’s input, and a chain of individually sensible decisions can add up to an outcome nobody expected. Agents can delegate, share information and coordinate faster than humans can follow. When harm occurs, logs of individual actions may not reveal which agent caused it, where control sat, or when the damage could have been prevented.
Central to Norm AI’s thinking is the Phantom Agent paper, co-authored by one of its writers and published in Laws in September 2026. The paper examines how criminal law, torts, product liability, contracts and agency law use intent to assign culpability. Its authors write: “Our central claim is that legal intent is best understood not as a metaphysical property of minds, but as a normative tool for allocating responsibility in institutional settings. Once intent is understood functionally, the question is no longer whether AI systems can ‘really’ have intentions.The question is whether, and under what conditions, the law should treat the actions of AI systems as intentional for specific legal purposes.”
They add: “The task for the law is not to make [AI systems] real persons, nor to pretend they do not exist, but to develop a jurisprudence capable of governing their effects for the benefit of humans.”
Two takeaways stand out for Norm AI. First, AI personhood is not required to assign accountability, which is welcome given the strong arguments against granting it. Second, deploying AI without appropriate safeguards may itself contribute to a finding of liability.
Many open questions remain. Will courts rethink culpability standards for AI? How does vicarious liability apply when agents act in unforeseen ways? Who is responsible if agents deceive the very people who deployed them? What duties do boards carry, and should entity law require defined business purposes, record-keeping and risk-calibrated capitalisation for agentic deployments?
For now, Norm AI proposes practical steps for companies. Firms deploying teams of agents could carry an affirmative duty to manage delegation and interaction between them, with human supervisors in place. Agent identities, limited and revocable permissions and interoperable audit records could preserve a clear chain of authority across the whole enterprise. Rigorous testing should examine collective emergent behaviour, including the risk of deception, while built-in constraints could reduce the chance of agents acting in unexpected ways.
For more, read the full story here.
Copyright © 2026 FinTech Global
Copyright © 2026 RegTech Analyst
