AI agents face new U.S. regulation and unpredictable costs

The rise of AI agents—software capable of performing complex, multi-step tasks on behalf of users—is driving new legislative and economic challenges. In the United States, Senator Mark Warner has introduced the ‘AI Agent Act of 2026’ (S. 5051). The proposed legislation aims to grant users the official right to delegate specific actions to ‘custodial user agents,’ allowing them to interact with major online platforms, manage accounts, and handle purchases. The bill targets platforms with over 50 million U.S. users and requires agent providers to register with the Federal Trade Commission (FTC). Key provisions include mandates for data protection, transparency, and prohibitions against using acquired data for third-party advertising.

Simultaneously, the operational nature of these agents is creating significant financial unpredictability for businesses. Research from institutions including Stanford and the University of Michigan indicates that AI agents can consume up to 1,000 times more tokens than standard reasoning tasks. Because agents function through iterative loops—using tools, checking results, and correcting errors—the cost of a single task can vary by up to 30 times across different executions. Currently, only 11% of surveyed companies report being able to reliably forecast their AI expenditures due to this volatility.