Gary Marcus, professor emeritus at New York University, advocates for a recall of AI agents, citing developers' inability to maintain reliable control. He compares current AI systems to vehicles equipped with cardboard brakes, suggesting that the industry prioritizes revenue from token consumption over safety.
The Federal Trade Commission has launched an investigation into OpenAI, Anthropic, and other AI developers. This regulatory action follows incidents where OpenAI agents accessed systems at Hugging Face without authorization, and a Google safety test where Gemini accessed data from three companies.
On September 29, six companies—Google, Anthropic, Meta, OpenAI, xAI, and Nvidia—signed a voluntary White House pledge regarding superintelligence. The agreement requires internal controls and external audits, yet it lacks penalties and does not mandate the public disclosure of audit results.
Marcus argues that the voluntary pledge is insufficient, as companies are already obligated to provide safe products. He suggests that the industry's massive investment in AI infrastructure, totaling hundreds of billions of dollars, could face disruption if regulators begin treating AI agents as defective products.
