Consumer Privacy Law Amendments for LMAs

Determine the specific amendments and tailoring points within existing U.S. consumer privacy laws—such as the California Consumer Privacy Act—that would effectively address the distinct challenges posed by language model agents (LMAs), including their uses of personal data for training and operation and the need for real-time informed consent mechanisms.

Background

The paper argues that regulating personal data could be an effective, indirect strategy to mitigate risks from platform-controlled AI agents, given the centrality of such data to training and operating LLM agents (LMAs). It notes potential measures such as imposing strict controls on personal data use for LMA training/operation and requiring real-time informed consent.

While the U.S. lacks broadly applicable federal consumer privacy laws, several large states (e.g., California) have enacted significant privacy statutes. The authors highlight uncertainty about how these laws should be amended or tailored to address LMAs’ distinct challenges, framing this as an open question that invites legal and policy analysis.