AI Critic Gary Marcus Warns ‘Reckless’ AI Could Cost Lives, Calls for Regulation

Cognitive scientist and longtime artificial intelligence skeptic Gary Marcus is again sounding the alarm about the pace of AI deployment, warning in a Fox Business interview that what he describes as “reckless” development of the technology could ultimately lead to deaths — and that governments have yet to put meaningful guardrails in place.

Marcus, an emeritus professor at New York University and the founder of two AI companies, has spent years positioning himself as one of the industry’s most persistent in-house critics. While many researchers debate distant, speculative risks from superintelligent machines, Marcus has tended to focus on nearer-term harms: systems that confidently generate false information, chatbots that give unreliable advice, and automated tools rolled out into sensitive settings before anyone has established whether they actually work.

His latest warning follows that same pattern. The concern is not that machines will suddenly develop intentions of their own, but that companies racing one another to ship products will put unreliable systems into contexts where errors carry real consequences — medical guidance, mental health support, vehicles, infrastructure, and the broader information environment that shapes public decisions.

A familiar argument, a louder moment

Marcus rose to wider prominence in 2023, when he testified before a U.S. Senate subcommittee alongside OpenAI chief executive Sam Altman and called for new oversight mechanisms for AI. In the years since, he has argued that the gap between what AI companies promise and what their systems reliably deliver remains wide — and that large language models, for all their fluency, still lack dependable reasoning and factual grounding.

That gap is precisely what makes rapid deployment dangerous, in his telling. A system that is right most of the time can still be catastrophic in the cases where it fails, particularly if users have been encouraged to trust it and no human remains meaningfully in the loop.

The regulatory vacuum

Marcus’s call for regulation lands at a moment when policymakers in the United States remain divided over how — and whether — to constrain the industry. The European Union has moved ahead with its AI Act, a risk-tiered framework that imposes stricter obligations on systems used in high-stakes domains. In the U.S., oversight has been patchier, assembled from executive actions, agency guidance, and a growing patchwork of state laws, while federal legislation has repeatedly stalled amid lobbying and disagreement over preemption.

Industry leaders frequently argue that heavy-handed rules would slow innovation and cede ground to international competitors. Critics like Marcus counter that voluntary commitments have limited enforcement value, and that waiting for a serious incident before acting is the costliest possible approach.

What regulation might look like

Proposals circulating among researchers and lawmakers include pre-deployment safety testing for high-risk applications, independent auditing, transparency requirements about training data and known failure modes, clear liability rules when automated systems cause harm, and labeling obligations for synthetic media. None of these ideas is new, but none has been broadly adopted in the U.S. either.

Marcus’s intervention is unlikely to change that overnight. But it adds to a growing chorus — including some figures inside the industry — arguing that the current trajectory, in which capability announcements consistently outpace safety infrastructure, is not sustainable.

Whether that argument gains traction may depend less on expert warnings than on the first high-profile failure that policymakers cannot ignore. That, Marcus has suggested, is exactly the outcome regulation should be designed to prevent. Read More