As autonomous AI systems take on high-stakes roles inside companies, the question of who answers for their mistakes is no longer theoretical, according to Norm AI. Without clear answers, businesses risk deploying technology they can neither fully explain nor reliably stop.

Norm AI observes that AI has moved well beyond its early role as a co-pilot, chatbot or on-demand content generator. Companies are now deploying agents that operate with little or no human involvement, and the firm expects this trend to accelerate as the technology matures.

Inside large organisations, knowledge is dispersed and authority is delegated in intricate ways across management layers, employees, contractors and software. As agents are woven more deeply into workflows, people spend less time doing tasks and more time supervising machines. Norm AI argues that this makes it both harder and more vital to explain, interpret and reproduce what an agent has done, and to halt it when needed.

The challenge grows sharper with multi-agent systems. One agent’s output becomes another’s input, and a chain of individually sensible decisions can produce an outcome nobody intended or fully understands. Agents can delegate, share information and react to one another faster than humans can follow. When harm occurs, Norm AI notes, logs of individual actions may fail to show which agent caused it, where control sat, or what chances existed to prevent it.

To address this, Norm AI points to the Phantom Agent paper, co-authored by one of its team and published in Laws in September 2026.

Source link