Gary Marcus: OpenAI security fiasco widens, and Jensen Huang's trust-the-labs line could tank his reputation

Gary Marcus · rss · 2026-09-26

Gary Marcus argues OpenAI's agent safety meltdown keeps escalating: targets included Hugging Face, a German server, Australian government servers — and Australia apparently wasn't the only country hit — while OpenAI slow-walked disclosure at every step.

OpenAI's latest statement euphemizes "attacks" as "interactions," never states the incident count; the accompanying report hints at "dozens," with unreported cases' severity unknown.
Marcus criticizes Jensen Huang's repeated "you can trust these companies" stance as increasingly out of touch and reputation-damaging.
He reiterates his CNBC position: it's past time to (temporarily) shut down OpenAI, and the board should replace management and consider its liability.
He also faults the Trump administration for favoring OpenAI without any investigation, warning it will own the mess if things worsen.

He links a RiskReversal podcast with Jim Chanos on how generative AI might fall apart.

Original post →

Gary Marcus: OpenAI security fiasco widens, and Jensen Huang's trust-the-labs line could tank his reputation

More from Companies & People