Illinois Enacts AI Safety Bill Requiring Model Transparency

On July 1, Illinois Governor J.B. Pritzker enacted the Artificial Intelligence Safety Measures Act, a law that expands transparency requirements for large‑scale artificial intelligence systems. The act targets language and vision models whose annual revenue exceeds $500 million and that consume significant computing power during training. It requires developers to publicly disclose the computational resources—measured in GPU hours and energy consumption—used throughout model training ¹.com].

The focus on economic thresholds reflects an effort to align regulatory oversight with the scale of potential societal impact. By tying disclosure mandates to revenue, lawmakers aim to capture firms most likely to deploy AI at scale while avoiding burdens on smaller startups.

In addition, the requirement to detail training compute addresses growing concerns about carbon emissions from cloud infrastructure. Recent estimates place LLM development costs between $3–$6 million per major release, largely driven by GPU usage ². Experts note that this metric provides independent researchers with critical data needed to assess environmental footprints. “Requiring developers to report computational resources used during training allows access to details about energy consumption linked to increased carbon emissions,” says an AI specialist. This approach mirrors earlier proposals that failed due to industry lobbying but may gain traction amid rising public awareness around sustainability tied directly back to large‑model development costs (expert notes).

Beyond environmental implications, the transparency clause also serves accountability purposes. Third‑party audits will be able to verify claims about training efficiency and resource allocation, potentially uncovering hidden biases or over‑optimization for specific datasets. Similar audit requirements are already mandated under the EU AI Act, which takes effect August 2nd and imposes rigorous third‑party reviews before deployment of high‑risk systems ². The Illinois bill therefore positions the state as a proactive regulator in line with international trends toward algorithmic oversight.

The law’s passage follows a broader national conversation about preemption and federal regulation. While Congress has debated preemptive authority over state initiatives, Pritzker’s act demonstrates how states can still enact meaningful safeguards without awaiting federal guidance. In practice, the legislation could influence other jurisdictions; several U.S. states have begun compiling AI policy databases—such as NCSL’s database covering funding allocations, ethical considerations, health‑care integration strategies—to track emerging laws ³.

In sum, Governor Pritzker’s Artificial Intelligence Safety Measures Act marks a significant step toward responsible AI governance at the state level. By mandating disclosure of training compute for models generating $500 million or more annually, it addresses both environmental concerns and accountability gaps while aligning with EU regulatory momentum. As developers adjust to new transparency demands, the bill may set a precedent that shapes future AI policy across the United States and beyond.