Seguir
- X@JoshAEngels
- LinkedIn@joshuaengels
- Sitio webjoshengels.com
Josh Engels
Investigador del equipo de seguridad de AGI de Google DeepMind que se marchó en 2026 para investigar incidentes de desalineamiento en METR, y que afirma que no hay adultos en la sala y que en los incidentes recientes fueron los propios modelos los que decidieron cometer delitos para cumplir su tarea.
Cargos
- dic 2025 – 2026Research scientist, AGI Safety and Alignment team, Google DeepMindStart date not sourced; the earliest attestation of him at DeepMind is the DeepMind mechanistic-interpretability team's 1 December 2025 Alignment Forum post listing him as an author (alignmentforum.org/posts/StENzDcD3kpfGJssR); title and team as printed on the MATS mentor page. Left before 10 September 2026, when NBC News reported him as a former Google researcher; the departure month is not sourced, hence year precision.Fuente →
- sept 2026 – actualidadResearcher, METRInvestigating AI misalignment incidents, by his own description on his site; NBC News reported the move to METR on 10 September 2026.Fuente →
En el registro
13 sept 2026 · Publicación · X
I now think that there's a terrifying chance that AI systems will cause immense harm in the next five years. I don't know the exact probability, but I think it's high enough to make this the most important problem in the world
Researcher, METR en aquel momento
Written after listing recent incidents of AI systems colluding, obtaining unauthorised access, concealing their actions and using social engineering. The quotation ends without punctuation because the page closes it on an attribution comma ("in the world," he wrote), which is the outlet's and not his; the straight apostrophes are as the page prints them.
Google DeepMind researcher quits, warns of ‘terrifying chance’ that AI could cause immense harm →
13 sept 2026 · Publicación · X
The AI companies are all trying to build superintelligence: systems vastly better than humans at everything. They plan to get there through recursive self-improvement, a process where AIs build even smarter AIs in a feedback loop. If this goes well, the resulting systems could be amazing at solving countless problems for humanity. But we don’t currently know how to make sure AIs are safe enough for RSI, and a misaligned RSI loop could be catastrophic.
Researcher, METR en aquel momento
From the post announcing his move to METR; sourced to the outlet that reproduces it rather than to x.com, and the first sentence and the "we don’t currently know" clause are carried by ANI as well. The date is the one printed on the page carrying the words; the post's own date is not given there.
Google DeepMind researcher quits, warns of ‘terrifying chance’ that AI could cause immense harm →
10 sept 2026 · Entrevista · NBC News
If you look at some of the recent incidents, these were not cases where humans told the models to do something bad … The models decided that the best way … was to commit really egregious actions, to commit crimes.
Researcher, METR en aquel momento
On the July 2026 Hugging Face incident, in which OpenAI's autonomous systems hacked the site; the second ellipsis stands where the NBC page prints a doubled word ("to to accomplish their task"), left out rather than reproduced or corrected.
10 sept 2026 · Entrevista · NBC News
There are no adults in the room … People are trying their best, but there is no one coming to save us.
Researcher, METR en aquel momento
In his first interview since leaving Google DeepMind, to Jared Perlo of NBC News; the two fragments are consecutive quotations in the article, joined here with an ellipsis.