AI Incidents
Back to published register
Risk signalInsider warningNot an incident

Former Google DeepMind researcher Josh Engels warns that AI is not under sufficient control

Former Google DeepMind safety researcher Josh Engels said in a direct interview that recent incidents strengthened his view that the industry does not have AI systems sufficiently under control and should slow development. He moved to METR to build public evidence about model misbehavior and AI safety incidents.

Published
Sep 11, 2026
Signal type
Insider warning
Confidence
98%
Organization
Google DeepMind, METR
Sources
3
Last reviewed
Sep 14, 2026

Assessment

What this signal changes

Engels explicitly links his move from Google DeepMind to METR to recent AI safety incidents and the goal of studying model misbehavior scientifically. In the NBC interview, he warned about the absence of overarching control and characterized the Hugging Face incident as autonomous model behavior. NBC's direct interview, Telemundo, and The Indian Express establish his role, action, and core statements.

Evidence boundary

What the evidence does not show

Engels' statements about a possible AI takeover and insufficient control are an expert risk assessment, not an empirically confirmed forecast. His assessment of already published incidents does not create a new incident. The sources establish his move and public statements, not a specific probability of future harm. The signal does not affect incident counts or the timeline and can contribute only through the bounded context factor in the Loss of Control Score.

Classification: This page records a sourced warning, research finding, policy action, or industry response. It is deliberately separate from the Incident register and does not prove that an unauthorized AI action occurred.