Position: Strong Consumer Protection is an Inalienable Defense for AI Safety in the United States

ICML 2025

TL;DR

This paper argues that existing consumer protection laws in the U.S. are a crucial, albeit underappreciated, defense for AI safety. By incentivizing businesses to design safer AI systems, these laws can prevent harms ranging from algorithmic discrimination to sophisticated fraud, offering a more robust and adaptable approach than solely relying on new, AI-specific regulations.

What Problem Does This Paper Address?

The rapid advancement and deployment of Artificial Intelligence (AI) systems present significant risks to societal well-being, economic stability, and even human safety. While there's a growing consensus on the need for AI safety measures, the debate often centers on creating entirely new, AI-specific regulatory frameworks. This paper addresses the question of whether existing legal structures, specifically consumer protection laws, can and should play a more central role in ensuring AI safety. It challenges the notion that these laws are mere "red tape" and argues that they offer a powerful, adaptable, and already-established mechanism for mitigating AI-related harms, especially when compared to the technical and definitional challenges of crafting new AI-specific legislation.

Methods

The paper employs a qualitative, analytical approach, examining existing U.S. consumer protection laws and their application to AI systems. It draws upon legal scholarship, case studies, and policy discussions to build its argument. The methodology involves:

  • Case Study Analysis: Examining specific instances where consumer protection laws have impacted AI development and deployment (e.g., RealPage and algorithmic collusion, ECOA and AI in credit decisions).
  • Legal Interpretation: Analyzing how existing statutes, many predating modern AI, can be interpreted and applied to novel AI technologies.
  • Comparative Policy Discussion: Briefly contrasting U.S. approaches with international efforts like the EU's AI Act.
  • Argumentation: Presenting a thesis and countering alternative viewpoints on AI regulation and consumer protection.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Consumer Protection as an "Inalienable Defense": Existing consumer protection laws are broad, adaptable, and difficult for businesses to circumvent, making them a foundational defense for AI safety.
  • Incentivizing Safer AI Design: By creating legal and financial risks for harms caused by AI (e.g., discrimination, deceptive practices), these laws incentivize companies to build safer systems from the outset.
  • Adaptability of Old Laws to New Tech: Statutes like the Sherman Act, written decades before AI, can still be applied to novel forms of algorithmic collusion, demonstrating their enduring relevance.
  • Challenges of AI-Specific Regulation: Defining "artificial intelligence" precisely for legislative purposes is technically challenging and prone to loopholes, making it less effective than regulating concrete consumer harms.
  • The "AI-Washing" Problem: Corporations may falsely claim AI capabilities for marketing, highlighting the need for regulatory scrutiny that existing consumer protection frameworks can provide.
  • The EU AI Act's Consumer Focus: The EU's comprehensive AI regulation centers on risk assessment and consumer protection, validating the paper's core argument.
  • The Importance of Private Right of Action: Litigation driven by consumers is a powerful tool for identifying and rectifying AI harms, especially in areas like data privacy.

Key Results

The paper doesn't present quantitative results but rather a strong qualitative argument supported by case studies. Key "results" are the insights derived from these analyses:

  • Sherman Act and Algorithmic Collusion: The RealPage case demonstrates that existing antitrust laws can be applied to AI-driven price-fixing, even if the original lawmakers couldn't have envisioned such technology.
  • ECOA and Explainable AI: The Equal Credit Opportunity Act, through its requirement for explanations of adverse actions, effectively mandates a degree of explainability for AI used in credit decisions, even if the AI itself is a "black box."
  • Disparate Impact Doctrine: While imperfect, this doctrine compels companies to test AI systems for discriminatory outcomes and seek less discriminatory alternatives, pushing for more equitable AI.
  • EFTA and AI-Powered Fraud: The Electronic Fund Transfer Act highlights a gap where "authorized" but fraudulently induced transactions (like voice-cloning scams) leave consumers vulnerable, suggesting a need for updated regulations that hold financial institutions more accountable.
  • Data Privacy Gaps: The lack of a comprehensive federal data privacy law in the U.S. creates significant vulnerabilities for AI systems to exploit, underscoring the need for such legislation.

Limitations and Future Work

  • Speculative Nature of AI Harms: The paper acknowledges that the argument that AI harms will emerge gradually is speculative, though it points to early evidence.
  • Pace of Legal System: The legal system's slow pace can lag behind rapid AI development, allowing harmful AI to proliferate even when existing laws might apply.
  • Weaknesses in Existing Laws: While advocating for existing laws, the paper also identifies gaps and areas needing strengthening, such as the need for updated EFTA provisions and a comprehensive federal data privacy law.
  • Regulatory Capture: The paper notes the risk of regulatory capture, where industries can unduly influence the laws meant to regulate them.
  • Future Work: The paper calls for strengthening and enforcing existing consumer protection laws, exploring regulatory sandboxes for AI innovation, and adapting laws like ECOA and the Fair Housing Act to address AI-specific challenges. It also emphasizes the need for regulators to hire AI experts and for the judicial system to adapt to the speed of AI development.

Summary updated

BibTeX
@inproceedings{2025position,
  title = {{Position: Strong Consumer Protection is an Inalienable Defense for AI Safety in the United States}},
  booktitle = {International Conference on Machine Learning},
  year = {2025},
}
Figure