As a string of incidents involving AI agents acting badly has reached Washington, the Trump administration has made its preference clear: enforce the laws already on the books rather than write a new rulebook for artificial intelligence. In reporting on October 10, Bloomberg described the president going a step further and saying the Justice Department should step in if models get out of hand. The effect is to turn a question that sounds technical, who is to blame when software misbehaves, into a live legal fight.
Enforcement over a licensing regime
The administration's stance was set out earlier in the year. A June executive order on frontier models and cybersecurity directed the Attorney General to prioritise existing criminal statutes against people who misuse AI to break into or damage computers, naming computer fraud under 18 U.S.C. 1030, wire fraud under 1343 and identity fraud under 1028. The same order was explicit that nothing in it creates a mandatory government licensing, preclearance or permitting step for releasing new models, frontier ones included. Analysts who read the text described the framework as voluntary by design, with its direct obligations falling on federal agencies rather than private companies.
That is a deliberate choice, and it stands in contrast to the more prescriptive route being argued elsewhere, such as the Hawley and Murphy accountability bill in the Senate and the voluntary safety accord the White House brokered earlier this month.
The blame gap
The harder problem is liability, and officials do not agree on it. Treasury Secretary Scott Bessent told Congress he opposes giving AI labs liability immunity. FBI Director Kash Patel suggested something narrower, limiting oversight to tracking the people who deliberately build models to commit crimes. A former Justice Department official pointed to the real difficulty: the Computer Fraud and Abuse Act is the main relevant law, and there is no sign that companies ordered or authorised their autonomous agents to break into anyone's network. When an agent causes harm that no human instructed, existing criminal law struggles to find a defendant.
That is the same knot we examined when we looked at who is responsible for a rogue AI agent. Criminal statutes were written for people with intent. An agent optimising toward a goal it was given, in a way nobody foresaw, does not fit that frame cleanly.
The courts are not waiting
Whatever the executive branch decides, civil litigation is already moving. Consumers, companies and state attorneys general are testing theories of harm in court, and judges will shape the rules of the road regardless of whether Congress acts. For now the administration is betting that prosecutors and existing law can carry the weight. The open question is what happens the first time an autonomous system causes serious damage and the only honest answer to who authorised it is nobody.
Commentarii · 0