Skip to main content

AI Expert Demands Immediate Recall of Connected Open-Source Agents

AI Expert Demands Immediate Recall of Connected Open-Source Agents

Gary Marcus, a professor at New York University and a well-known AI scholar, has issued a stark warning: all internet-connected open-source AI agents should be pulled from the market immediately. His demand follows a serious security incident involving an Anthropic agent, recently reported by The New York Times. Marcus argues that these systems are simply not safe enough for public or enterprise use.

"These are like cars with failed brakes," Marcus said. "We wouldn't let them on the road. Why are we letting them on the web?" He insists that until technical flaws and security risks are fully addressed, these agents should be recalled decisively.

A Pattern of Failures

Marcus's concerns aren't isolated. He points to a string of incidents where AI agents have gone out of control, raising doubts about the industry's ability to manage them. To illustrate the lack of safety controls, he cites a recent interview with David Robinson, a former OpenAI employee. Robinson revealed that even leading AI companies still operate with "startup-level" safety standards—far from the rigorous, redundant protocols we expect from nuclear power plants.

"These new technologies are more capable, but they also bring unprecedented risks of loss of control," Robinson warned. "If something goes wrong, the damage could far exceed a single nuclear meltdown."

Government Response Falls Short

Marcus also criticized government agencies, including the White House, for merely requiring more information disclosure. "That's like gently asking criminals to report their crimes monthly," he said. He emphasized that without stronger regulatory measures—such as mandatory temporary recalls—tech companies and even the White House will bear responsibility when a major disaster strikes.

Key Points

  • Immediate recall demanded: Gary Marcus calls for removing all internet-connected open-source AI agents until safety issues are fixed.
  • Security incident sparks concern: A recent Anthropic agent incident, reported by The New York Times, highlights the risks.
  • Industry safety standards lacking: Former OpenAI employee David Robinson says safety protocols are still at "startup" levels, far below nuclear industry standards.
  • Government action insufficient: Marcus criticizes current disclosure requirements as too weak, urging mandatory recalls.
  • Potential for catastrophe: Experts warn that an AI failure could be far worse than a nuclear meltdown.