FTC is investigating OpenAI, Anthropic and other AI companies over product risks

The FTC's regulatory authority over technology companies stems from a foundational statute passed in 1914 that gives the agency broad power to investigate unfair or deceptive practices in commerce. For most of its history, the FTC applied this framework to conventional consumer harms: false advertising claims, price fixing, privacy violations. But as AI systems have become more integrated into products that touch millions of people, regulators face a novel challenge. The existing legal toolkit was designed for predictable, measurable harms. AI introduces uncertainty at scale. A company might not know exactly how its system will behave in all contexts, making it difficult for regulators to pin down whether something is actually deceptive or just unpredictable. The concept of "product risk" in AI is itself evolving. In traditional consumer products, risk is often quantifiable. A car maker tests for crash safety. A pharmaceutical company runs trials to measure adverse effects. With large language models and other AI systems, the failure modes are harder to isolate. A model might generate harmful content, exhibit biases, hallucinate false information, or behave differently depending on how a user phrases a question. These aren't necessarily bugs in the conventional sense. They can emerge from the nature of how these systems learn from data. Regulators must decide whether a company has done enough to understand and disclose these behaviors before releasing a product to the public. There is genuine historical precedent for regulators moving into unfamiliar territory. When automobiles first became common, traffic deaths spiked and the legal system had to adapt. When pharmaceuticals expanded, the FDA emerged to require safety testing before market entry. Environmental regulation followed chemical spills and contamination. In each case, regulations developed after real harms were observed, often after people were injured. The current regulatory moment around AI is unusual because agencies are attempting to establish frameworks before widespread harm has crystallized into established patterns. That requires regulators to think probabilistically rather than reactively. One central tension is that comprehensive testing and safety work takes time and money. A small startup building AI tools faces different constraints than an established tech company with dedicated research staff. Regulators must balance the impulse to ensure safety with the risk of creating barriers so high that only the largest companies can operate. The history of regulation shows this is not a problem with an obvious answer. Overly strict early rules can stifle innovation and competition. Insufficient oversight can allow preventable harms to compound. Another layer of complexity is transparency. Companies generally have strong incentives not to publicize vulnerabilities or failure modes in their products, fearing liability or competitive disadvantage. Yet regulators and inde