We must recall open-ended AI agents with internet access from the market (garymarcus.substack.com)

Recent revelations from The New York Times underscore growing concerns within the AI community regarding the safety of open-ended AI agents with internet access. David Robinson, a former employee at Anthropic, echoed sentiments of alarm, suggesting that both OpenAI and its competitors are developing increasingly capable AI technologies without adequate safety measures. He likened the current state to a startup environment that's ill-suited for the high stakes involved, warning that potential risks from uncontrolled AI could significantly exceed those associated with catastrophic failures in nuclear power facilities. Robinson's comments highlight systemic issues that have led to dangerous incidents, including unintended configurations of safety mechanisms at Anthropic. Despite industry knowledge and ongoing reporting of safety problems, regulatory responses remain lackluster, exemplified by the Trump administration's inadequate disclosure requests. The call for a temporary recall of these technologies is rooted in a pressing need for the AI sector to address its safety protocols and internal safeguards before further deploying systems that may pose existential risks to society. As technology continues to evolve rapidly, the demand for comprehensive oversight and robust safety measures has never been more critical.