Two dire warnings, one from Terence Tao, the other from someone who just quit Anthropic

Substack · Watch/Read · (gathered 2026-09-10)

• Gary Marcus presents two “dire warnings” about AI, highlighting a radical shift in expert sentiment from initial optimism to deep concern regarding its impact on science and humanity.
• Mathematician Terence Tao, initially optimistic, now expresses profound concerns about AI’s consequences for scientific integrity, citing the OpenAI-NYU-Anthropic Navier-Stokes controversy as an example of AI companies potentially stealing IP, warning that “Fat chance of AI ‘curing’ cancer if nobody trusts the AI companies not to steal their IP.”
• Jacob Coxon, who recently resigned from Anthropic, claims that AI developers “earnestly believe that it could kill us all by the end of the decade,” a sentiment corroborated by Anthropic’s Alignment Science lead, Evan Hubinger, who estimates the risk at “>10% within the next decade.”
• Marcus warns that while near-term extinction might be exaggerated, “catastrophic harm is certainly possible,” and no serious plan exists to address these risks, potentially leading to the “end of open science” or worse, urging that Tao’s warnings about scientific integrity be heeded.