This AI Company Wants Washington To Keep Its Competitors Off the Market

Dario Amodei, CEO of the artificial intelligence company Anthropic, published a guest essay in The New York Times Thursday arguing against a proposed 10-year moratorium on state AI regulation. Amodei argues that a patchwork of regulations would be better than no regulation whatsoever.

Skepticism is warranted whenever the head of an incumbent firm calls for more regulation, and this case is no different. If Amodei gets his way, Anthropic would face less competition—to the detriment of AI innovation, AI security, and the consumer.

Amodei’s op-ed came in a response to a provision of the so-called One Big Beautiful Bill Act, which would prevent any states, cities, and counties from enforcing any regulation that specifically targets AI models, AI systems, or automated decision systems for 10 years. Senate Republicans have amended the clause from a simple requirement to a condition for receiving federal broadband funds, in order to comply with the Byrd Rule, which in Politico‘s words “blocks anything but budgetary issues from inclusion in reconciliation.”

Amodei begins by describing how, in a recent stress test conducted at his company, a chatbot threatened an experimenter to forward evidence of his adultery to his wife unless he withdrew plans to shut the AI down. The CEO also raises more tangible concerns, such as reports that a version of Google’s Gemini model is “approaching a point where it could help people carry out cyberattacks.”

Matthew Mittelsteadt, a technology fellow at the Cato Institute, tells Reason that the stress test was “very contrived” and that “there are no AI systems where you must prompt it to turn it off.” You can just turn it off. He also acknowledges that, while there is “a real cybersecurity danger [of] AI being used to spot and exploit cyber-vulnerabilities, it can also be used to spot and patch” them.

Outside of cyberspace and in, well, actual space, Amodei sounds the alarm that AI could acquire the ability “to produce biological and other weapons.” But there’s nothing new about that: Knowledge and reasoning, organic or artificial—ultimately wielded by people in either case—can be used to cause problems as well as to solve them. An AI that can model three-dimensional protein structures to create cures for previously untreatable diseases can also create virulent, lethal pathogens.

Amodei recognizes the double-edged nature of AI and says voluntary model evaluation and publication are insufficient to ensure that benefits outweigh costs. Instead of a 10-year moratorium, Amodei calls on the White House and Congress to work together on a transparency standard for AI companies. In lieu of federal testing standards, Amodei says state laws should pick up the slack without being “overly prescriptive or burdensome.” But that caveat is exactly the kind of wishful thinking Amodei indicts proponents of the moratorium for: Not only would 50 state transparency laws be burdensome, says Mittelsteadt, but they could “actually make models less legible.”

Neil Chilson of the Abundance Institute also inveighed against Amodei’s call for state-level regulation, which is much more onerous than Amodei suggests. “The leading state proposals…include audit requirements, algorithmic assessments, consumer disclosures, and some even have criminal penalties,” Chilson tweeted, so “the real debate isn’t ‘transparency vs. nothing,’ but ‘transparency-only federal floor vs. intrusive state regimes with audits, liability, and even criminal sanctions.’”

Mittelsteadt thinks national transparency regulation is “absolutely the way to go.” But how the U.S. chooses to regulate AI might not have much bearing on Skynet-doomsday scenarios, because, while America leads the way in AI, it’s not the only player in the game. “If bad actors abroad create Amodei’s theoretical ‘kill everyone bot,’ no [American] law will matter,” says Mittelsteadt. But such a law can “stand in the way of good actors using these tools for defense.”

Amodei is not the only CEO of a leading AI company to call for regulation. In 2023, Sam Altman, co-founder and then-CEO of Open AI, called on lawmakers to consider “intergovernmental oversight mechanisms and standard-setting” of AI. In both cases and in any others that come along, the public should beware of calls for AI regulation that will foreclose market entry, protect incumbent firms’ profits from being bid away by competitors, and reduce the incentives to maintain market share the benign way: through innovation and product differentiation.

The post This AI Company Wants Washington To Keep Its Competitors Off the Market appeared first on Reason.com.