A bipartisan coalition of lawmakers introduced sweeping legislation on Friday that would, for the first time, establish a comprehensive set of federal rules governing artificial intelligence, a move that supporters described as the most consequential attempt yet to bring order to a technology that has raced far ahead of the law. The measure, titled the Responsible Artificial Intelligence Act, would cover everything from hiring software to medical diagnostic tools.
At the heart of the bill is a tiered system that sorts automated systems by the risk they pose. The highest tier, reserved for applications that can affect someone's access to credit, housing, employment or medical care, would face the strictest requirements, including independent audits, ongoing monitoring and a public registry of where the systems are deployed.
“For too long, the people most affected by these systems have had the least say in how they work,” said Representative Dana Whitfield, one of the bill's lead authors, at a news conference on the Capitol steps. “This framework is not about slowing innovation. It is about making sure that when a machine makes a decision about your life, you have the right to know how, and the right to challenge it.”
The legislation would create a new Office of Algorithmic Accountability, housed within the Commerce Department, to certify auditors, investigate complaints and levy fines on companies that deploy unreviewed high-risk systems. Firms found in repeated violation could face penalties of up to four percent of their annual revenue, a figure modeled on the data-privacy laws of several allied nations.
Industry reaction was swift and divided. The Coalition for Open Innovation, a trade group representing several large software developers, warned that the audit requirements could entrench incumbents by burdening smaller firms with compliance costs they cannot absorb. “We support sensible guardrails,” the group said in a statement, “but a one-size-fits-all mandate risks freezing out the very startups that drive progress.”
Civil-liberties advocates, by contrast, argued the bill did not go far enough. Researchers at the Lattice Institute, an independent policy center, noted that the proposal exempts most systems used by law enforcement and leaves open questions about how audits would handle models that are continuously retrained. “A snapshot audit tells you little about a system that changes every week,” said Priya Venkatesan, the institute's director of technology policy.
The bill's path forward remains uncertain. Committee hearings are expected to begin next month, and aides on both sides acknowledged that the disclosure provisions, in particular, could be narrowed before any vote. Several lawmakers have signaled they will push to add carve-outs for academic research and open-source projects, fearing that broad rules could chill experimentation.
Even so, the framework's arrival marks a turning point in a debate that has, until now, played out largely in voluntary commitments and patchwork state laws. Whether it becomes the foundation of a durable national standard or stalls amid competing interests, its sponsors say, the conversation about who answers for the decisions machines make has finally reached the floor of Congress.