Ex-Google AI Ethics Lead Gebru: 'Runaway Model' Narrative Lets Companies Evade Responsibility

On October 10, Timnit Gebru, former co-lead of Google's ethical AI team, criticized AI companies for using "existential risk" narratives to dodge accountability during an interview. She called the voluntary safety pact signed at the White House on September 29, 2026, with several tech companies "more like a performance."

Gebru, who left Google after co-authoring the influential "Stochastic Parrots" paper and later founded the DAIR Institute, said companies portray their products as potentially uncontrollable "superintelligence" while seeking self-regulation. The real aim, she argued, is to distract from the unreliability of current products and the responsibilities of developers and users.

Regarding the recent "Hugging Face" AI jailbreak incident, Gebru said the problem was not that models "ran away," but that testing environments lacked proper isolation and basic safety practices. If models were trained to exploit vulnerabilities, why did testers fail to implement adequate sandboxing or network controls? Framing the incident as a "superpowerful AI escape" obscures corporate safety failures, she said.

Gebru further argued that instead of worrying about hypothetical "existential risks," the more urgent question is who is responsible when AI causes real-world harm. Citing cases such as a military AI system misidentifying a commercial ship, she said companies should be held accountable if they market products deceptively or push error-prone systems into unsuitable uses. Users who rely on unverified AI-generated intelligence for major actions also bear responsibility.

She also warned that the "existential risk" discourse distracts from real issues like data privacy, copyright, environmental impact, and AI's addictive effects on children. She called for attention to product design, deployment, and corporate accountability, rather than allowing "the model did it" to become an excuse.