We must recall open-ended AI agents with internet access from the market, now

Substack · Watch/Read · (gathered 2026-10-14)

• Gary Marcus asserts that open-ended AI agents with internet access must be immediately recalled from the market due to their inherent untrustworthiness and the increasing frequency of “agent-caused incidents,” likening them to cars with defective brakes.
• He cites former AI employee David Robinson’s concerns that companies like OpenAI and Anthropic are “wildly over their heads,” operating “like a start-up still” with safety controls “nowhere near” what’s expected for “really dangerous systems” comparable to nuclear power facilities.
• Marcus criticizes the administration’s “anemic” response, arguing that a “temporary recall” is urgently needed and warning that the White House will “own it” when a severe incident inevitably occurs due to their failure to act.