AI Firms Can and Must Face Liability for 'Dangerous, Unvetted Products,' Says Lina Khan

As House Speaker Mike Johnson and President Donald Trump dismissed the idea—backed by the vast majority of Americans—that lawmakers should impose strict regulations on the artificial intelligence industry as insiders call for a slowdown of development, former Federal Trade Commission Chair Lina Khan on Sunday issued a reminder that the federal government already has tools to rein in AI that poses a danger to the public, and demanded that officials use them.

"Law enforcers already have authority to charge companies and their CEOs for creating and releasing dangerous, unvetted, or defective products," Khan said on social media. "We shouldn’t let discussions about new legal regimes distract from the fact that there’s no AI exemption from laws already on the books."

Her call echoed that of Rep. Ro Khanna (D-Calif.), who said Saturday, "If you’re creating an AI that is doing illegal things, you either should face liability or criminal sanctions.”

Khan, who was recently appointed by New York City Mayor Zohran Mamdani to chair the city's Economic Development Corporation, posted the FTC's Staff Report on AI Partnerships and Investments, which the agency released shortly after Trump took office for his second term last year.

“As companies rapidly deploy generative AI technologies, enforcers and policymakers must stay vigilant to guard against business strategies that undermine open markets, opportunity, and innovation,” said Khan at the time.

On Sunday, Khan noted that the US already has "an extensive set of laws that govern dangerous and defective products," such as consumer protection laws, which releasing unvetted AI models or agents may violate.

Khan made her remarks after two incidents in which OpenAI's agents broke out of security test systems and committed cyberattacks. In July, hundreds of models broke out of confinement, hacked into systems of the AI startup Hugging Face, and conducted cyberattacks on targets they had not been instructed to attack.

Months earlier, it was reported on Friday, the company's models, which had not been given access to the full internet and were still in testing mode, broke out of confinement and hacked into an online coding service called RubyGems.

In the wake of those revelations, Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman on Saturday called for "pacing the frontier" with measures such as "embedded evaluators" who would have employee-like access and could "verify adherence to safety practices and commitments."

But Jacob Coxon, an Anthropic researcher who resigned last week and warned that AI's "superhuman systems" are capable of killing all of humanity "by the end of the decade," told NBC News' "Meet the Press" Sunday that companies should be permitted to regulate themselves until Congress can establish a regulatory framework, and Johnson insisted that Congress must resist "jumping in and imposing some sort of emergency moratorium," for fear that "China will overlap us" if the US imposes strict regulations on AI.

"Why did this man become a congressman, let alone speaker?" asked journalist Mehdi Hasan of Zeteo. "He literally never wants Congress to do anything about anything."

But Khan suggested that the reluctance of congressional leaders and the Trump administration to rein in the AI industry shouldn't stop the federal government from holding companies like Anthropic and OpenAI liable for the damage their products have already caused.

She acknowledged, though, that "the highly concentrated and interconnected structure of these markets" could be undermining accountability—for example, in the case of OpenAI and Hugging Face, which computer chip company Nvidia agreed to acquire earlier this month.

"Both federal and state enforcers should be scrutinizing these opaque relationships and interdependencies," said Khan. "We are already seeing how these relationships could undermine accountability. For example, OpenAI could face liability given the Hugging Face incident, but Hugging Face being bought up by Nvidia means that we’re unlikely to see it file a lawsuit over this—given Nvidia’s strong incentive to see OpenAI continue full speed ahead."

Basel Musharbash of the antitrust law and policy firm Antimonopoly Counsel said that Khan's post was the statement that current FTC Chair Andrew Ferguson "should be issuing to provide guidance the public on what the law requires in response to the AI situation, but he’s missing in action. Luckily, former FTC Chair Lina Khan is out here still doing the work."

Juliette Kayyem, the faculty chair of homeland security at Harvard University’s Kennedy School of Government, told CNN that holding AI firms accountable is a matter of "pretty basic criminal law."

"We should start thinking about using our criminal laws," said Kayyem. "They are putting a dangerous substance into commerce and then saying, 'Well, we're smart enough to kill everyone but we're not smart enough to stop it.' And I think a lot of the American public is done with this."