AI CEOs Want Washington to Act, but Congress Still Has No Common Rulebook

WASHINGTON — Some of the artificial-intelligence industry's most prominent executives are urging the federal government to build stronger oversight, creating a rare situation in which leaders of a fast-growing technology sector are publicly asking Washington for rules. The appeal has not produced agreement in Congress. Lawmakers remain divided over whether the immediate priority should be safety, competition with China, worker protection, national security or preserving room for American companies to move quickly.

The debate intensified after leaders associated with Anthropic, OpenAI and other major AI companies warned that increasingly capable systems require a more deliberate approach. Their proposals are not identical, and support for oversight should not be mistaken for a shared legislative blueprint. Companies can agree that regulation is necessary while disagreeing sharply over who conducts testing, what triggers a reporting duty and whether any rule slows the release of a product.

Photo: Architect of the Capitol · Public domain

That distinction matters for business. A federal framework could reduce the uncertainty created by separate state laws, but poorly designed rules could also favor the largest companies. Frontier-model developers have the lawyers, computing budgets and compliance teams to absorb expensive testing regimes. Startups may not. Congress has to decide how to impose meaningful safeguards without turning compliance costs into a protective wall around today's market leaders.

The political split is unusually complicated. Some lawmakers want mandatory safety evaluations, incident reporting and limits on the most powerful systems. Others argue that an emergency pause would allow foreign rivals to close the gap. President Donald Trump has rejected calls for a slowdown, while members of both parties have proposed narrower measures involving national security, civil liberties, labor and catastrophic-risk monitoring. The result is concern without a common theory of action.

Senator Josh Hawley has introduced bipartisan legislation aimed at federal tracking of certain AI safety concerns and separately opened an investigation into a reported cybersecurity evaluation incident. Senator Bernie Sanders and Representative Greg Casar have proposed far more aggressive limits on artificial superintelligence. These efforts show that Congress is not ignoring the subject, but they also reveal how far apart the possible endpoints remain—from disclosure and monitoring to an outright prohibition on a class of development.

The White House and congressional leaders are also weighing economic stakes. AI investment is supporting data-center construction, chip demand, software spending and company valuations. A sudden constraint would move through supply chains and capital markets. At the same time, a serious safety failure could destroy trust, invite litigation and produce a more chaotic policy response later. The commercial question is not regulation versus growth; it is which rules make growth durable enough to survive public scrutiny.

States are already filling parts of the vacuum. California has enacted new child-safety obligations for social media and companion chatbots, while regulators are examining automated pricing and data use in other contexts. BizzNews readers can also review the FTC's personalized-pricing proposal, which illustrates how existing consumer-protection law is being applied to algorithmic decisions. A patchwork may eventually push national companies toward a single internal standard even before Congress acts.

Businesses outside the AI industry should not wait for a sweeping federal law before preparing. Companies deploying models can document which systems they use, what data those systems receive, who reviews high-impact outputs and how incidents are escalated. Procurement contracts should identify responsibility for testing and security rather than assuming the vendor carries every risk. The fastest way to lose control is to discover after a failure that no team owns the decision.

Voluntary standards can help, but they have limits. Companies are capable of sharing technical practices and commissioning independent evaluations, yet self-regulation cannot supply public accountability or resolve conflicts of interest on its own. Antitrust rules may also complicate coordination among competitors. A credible system will probably require a combination of company testing, independent review and government authority, with thresholds that become stricter as models gain capabilities.

The most useful near-term legislation may be less dramatic than the public debate. Clear incident-reporting rules, protections for researchers, common evaluation methods and transparency around high-risk deployments could build an evidence base for later decisions. Congress does not need to settle every philosophical question about intelligence before improving oversight. It does need enough technical capacity to distinguish an enforceable safeguard from a slogan written for a hearing.

The industry's request for regulation should be treated as information, not proof that any particular proposal is correct. Executives know their systems well, but they also have commercial incentives and competing visions of the market. Lawmakers must test the details in public. The outcome will shape which companies can compete, how quickly products reach customers and who bears the cost when an AI system causes harm. That is a business-policy decision too large to leave to either side alone.

Related coverage: Google's court-ordered ad-tech business changes and the FTC's proposed disclosure rules for personalized pricing.