The Texas sponsor of one of the most closely watched artificial intelligence bills in the country has scrapped it in favor of a fresh proposal that no longer focuses on algorithmic discrimination in high-risk contexts.
Rep. Giovanni Capriglione’s (R) new bill, introduced Friday, instead targets AI systems more broadly and seeks to bar the technology from discriminating against protected classes, censoring political views or inciting criminality.
“While the language on this bill may be different or new, this bill is a better reflection of the stakeholder input that we have had from various groups over the past few months,” Spencer Ward, executive director of the Innovation and Technology Caucus of the Texas Legislature, said in an email.
Capriglione did not immediately respond to a request for comment.
The earlier bill, dubbed the Texas Responsible AI Governance Act, borrowed from a framework that emerged from a bipartisan multistate working group of lawmakers who have focused their efforts on addressing discrimination by AI systems in high-risk settings such as housing, employment and health care.
The first of those bills to become law was passed in Colorado last year. Similar measures have been introduced in more than a dozen states this year, including legislation in Virginia awaiting Gov. Glenn Youngkin’s signature.
Capriglione’s original bill was viewed as a potential red state model for regulating high-risk AI systems. While it still placed obligations on AI developers and deployers, it included a broad exemption for small businesses and exempted AI systems that are still being tested from regulations by creating a “sandbox” program.
Read more: Texas lawmaker unveils sweeping AI bill for 2025
But even with those business-friendly provisions, the legislation was pilloried by free market and libertarian groups who said it was a “blue-state model that hampers innovation.” Those criticisms gained more currency with the Trump administration’s relatively hands off approach to AI regulation and warnings that the U.S. is in an AI arms race with China.
Capriglione pushed back on the criticism last month in a message to Pluribus News: “From algorithmic biases to censorship concerns, there are valid reasons for Texans to demand accountability in the way AI impacts our economy and society.”
He also said he would continue to work with “stakeholders” as the legislative session progressed.
Capriglione’s new bill, which is still called the Texas Responsible AI Governance Act, no longer specifically addresses high-risk AI systems that are a “substantial factor” in consequential decisions about people’s lives.
Instead, the bill adopts a broader prohibition on the development of AI systems “with the intent to unlawfully discriminate against a protected class.” The bill language makes clear that disparate treatment alone is not sufficient to prove an intent to discriminate.
The bill also requires government agencies to disclose to the public if they are interacting with AI. And it adds a new section barring AI systems from being intentionally developed to “incite harm or criminality.”
Another element of the bill bars governments from using AI to engage in “social scoring” of individuals. The earlier bill didn’t limit it to government use.
Perhaps most significantly, the new bill adds language prohibiting AI systems from intentionally engaging in “political viewpoint discrimination.”
Read more: AI regulation advocates steer around more potholes
“It is surprising and disappointing to see Rep. Capriglione include provisions on content moderation that have more or less been rejected by the U.S. Supreme Court in his reframed approach to AI regulation,” said Scott Babwah Brennen, director of the Center on Technology Policy at New York University.
Conservatives have previously seized on content moderation by social media platforms as a form of censorship, resulting in a pair of 2021 laws in Florida and Texas that the tech industry challenged on First Amendment grounds in cases that reached the Supreme Court.
In his new bill, Capriglione retained prohibitions on using AI to create child sexual abuse material. It also still allows consumers to appeal decisions made by AI systems that have an adverse effect on their “health, welfare, safety, or fundamental rights.” And it retains the “sandbox” program.
Both bills also establish a Texas Artificial Intelligence Council tasked with ensuring that AI systems are “ethical and in the public’s best interest and do not harm public safety or undermine individual freedoms.” The council would also investigate cases of “undue influence” by technology companies or undue burdens on smaller developers.
The new legislation would still be enforced by the attorney general but does not include the broad exemption for small businesses that was included in the earlier bill.
Capriglione’s rewrite of his AI bill won immediate praise from Adam Thierer, a senior fellow at the libertarian R Street Institute, who had criticized the initial version.
“Capriglione’s revised bill represents a sensible course correction to steer Texas away from a European-style regulatory approach for AI,” Thierer said. “While the new bill’s content moderation regulatory provisions still raise some red flags, this bill now takes a more balanced approach to AI policy that is consistent with Texas values and the Trump administration’s approach to these issues.”
Thierer said Capriglione’s departure from the multistate working group’s model signals a new direction for AI regulation in Republican-led states.
Babwah Brennen of NYU said that while the bill includes new protections against intentional discrimination, “it ultimately offers a far more minimalist approach to AI safety and AI regulation” than Capriglione’s earlier version.
Including Capriglione’s bill, nearly two dozen other AI-related bills were introduced in the Texas legislature last week ahead of the Friday deadline to file legislation, according to a review by Babwah Brennen.
They include a measure from Rep. Hillary Hickland (R) that would require an AI review of books purchased by school districts to search for “sexually explicit” or “sexually relevant” content.