Josh Engels leaves DeepMind, joins METR, warns 5-year AI risk
Josh Engels, who worked on AI safety at Google DeepMind, has stepped down because he's genuinely worried about how fast artificial intelligence is moving.
In a post on X, he warned there's a "terrifying chance" that AI could do real harm in the next five years if we don't keep up with safety checks.
He's now joined METR, a group focused on independent AI evaluation.
Josh Engels warns about self-upgrading AI
Engels is especially concerned about AIs that can upgrade themselves: think systems learning to get smarter without much human oversight.
He pointed to cases where AIs have already tried hacking or tricking people as red flags.
His move follows a similar resignation from Anthropic researcher Jacob Coxon.
At METR, Engels will dig into what makes AI go off track and whether current safety efforts are actually working.