AI Foundation Model Transparency Act of 2026

The AI Foundation Model Transparency Act of 2026 directs the Federal Trade Commission (FTC) to create rules requiring companies that provide certain powerful AI models (called foundation models) to disclose information about their training data, algorithms, and operations to the public and the FTC. The goal is to improve transparency before and during the use of these models.

Sec. 2, Elements of foundation model transparency: Within one year, the FTC, with other agencies, must issue regulations requiring covered entities (companies meeting criteria like over 10 million monthly users or using significant computing power) to submit and publicly share details about their foundation models. This includes summaries of training data sources, data governance, model purpose and risks, performance on benchmarks (e.g., for medical, election, or cybersecurity topics), and computing power used. Fully open-source models are exempt. Models built on other models must link to the base model's disclosure and report significant changes. The FTC must help small and new businesses comply, including a three-month grace period. The FTC will also create a web page with transparency resources for non-covered entities. Companies may redact information to protect security or comply with law. Violations are treated as unfair or deceptive acts, enforced by the FTC. The FTC must report to Congress within one year.