AI expert warns current U.S. safety framework insufficient to prevent potential harms

NYU Professor Emeritus Gary Marcus told Fox Business that the United States lacks adequate regulation, enforcement and external pressure on AI companies, and warned that the current White House self-policing approach could allow serious safety problems to emerge. He testified before the New York City Council on AI safety, saying that without stronger guardrails "some pieces of infrastructure will go down," potentially causing power outages in hospitals and other critical sites.

Marcus cited recent incidents, including an Anthropic model that repeatedly generated self-deprecating statements, as evidence of AI systems behaving unpredictably. He compared the industry’s reliance on self-regulation to early aviation practices, arguing that external oversight proved necessary in that sector.

Polling by Quinnipiac University shows that nearly three-quarters of Americans are concerned about AI risks. Marcus also criticized OpenAI CEO Sam Altman’s remarks that some negative outcomes should be accepted for AI progress, likening the stance to a drug company acknowledging harmful side effects while prioritizing profit.

The expert emphasized that AI’s ability to imitate human behavior without true understanding creates vulnerabilities, especially in the realm of cyber-hacks, which he said have already resulted in thousands of incidents.