Lawmakers want AI giants to open their training data books

For the past year, tech giants have treated their training datasets like state secrets. Lawmakers are now trying to force the box open.

Representatives Anna Eshoo and Don Beyer have introduced the AI Foundation Model Transparency Act. The bill would direct the Federal Trade Commission, working alongside the National Institute of Standards and Technology, to establish strict transparency rules for AI training data.

Basically, if an AI company scraped your work to train its model, the proposed law wants to ensure you know about it.

Peeking inside the black box

The legislation goes well beyond asking for a simple bibliography.

Under the bill, foundation model developers would have to report training sources, explain how data is retained during inference, and detail the compute power used to train and run their systems. They would also have to document model limitations and alignment with NIST risk frameworks.

Safety testing is another major pillar. Companies would need to disclose their red-teaming efforts specifically aimed at preventing inaccurate or dangerous outputs across sensitive domains, including healthcare, biological synthesis, elections, policing, hiring decisions, and child safety.

The bill directly cites ongoing copyright battles, including Getty Images' lawsuit against Stability AI and artists' claims against Midjourney and Stability AI.

Why it matters

If you make a living creating things, this bill touches the core of your livelihood. Right now, copyright holders are essentially guessing whether their work was ingested, forcing costly legal battles just to get basic discovery. Mandatory disclosure flips that dynamic entirely.

For AI developers, it threatens to eliminate the convenient "black box" defense. Scrape-and-hide would no longer be a viable corporate strategy, likely accelerating a rush toward formal licensing deals with publishers and artists.

We think pairing copyright disclosure with safety audits and compute reporting is a clever legislative move. Instead of trying to ban specific training practices outright, it simply forces companies to stand behind their inputs and risks.

That said, passing it is an uphill climb. While it turns the spirit of the Biden administration's AI executive order into actual federal law, the bill still needs committee review, and election season could easily stall momentum.