Bilal Chughtai

Ingénieur de recherche dans l’équipe d’interprétabilité des modèles de langage de Google DeepMind, au sein de son équipe de sécurité et d’alignement de l’AGI, parti en juillet 2026 et aujourd’hui responsable d’un programme chez BlueDot Impact, qui affirme croire sincèrement que l’IA a le pouvoir de tous nous tuer, que l’alignement reste difficile et non résolu, et qu’il faut cadencer le développement à une vitesse que la société puisse supporter.

Fonctions

févr. 2025 – juil. 2026Research Engineer, language model interpretability team, Google DeepMindThe team sits inside DeepMind's AGI safety and alignment team, by his own description in the post announcing the job, dated 17 February 2025; he left in July 2026, as reported by The Next Web and Parameter, and his own site now lists the role as previous.Source →
sept. 2026 – aujourd’huiProgram Lead, BlueDot ImpactRunning a new AI safety upskilling programme, by his own description; the start date is not sourced, so this is the earliest attestation — the New York Post reported the move on 15 September 2026, citing Bloomberg.Source →

Au registre

14 sept. 2026 · Publication · X
We need to pace AI development to a speed that society can handle, where emerging risks can be addressed before extreme harm is realized.
Program Lead, BlueDot Impact à l’époque
His prescription, from the same post; Parameter's copy prints the British spelling "realised" where the New York Post prints "realized", and the word is left as the page cited prints it. The Next Web carries only the fragments "to avoid this manic race between AI companies" and "to a speed that society can handle" inside its reporter's own sentence, so the sentence is taken from the page that prints it whole.
Google DeepMind researcher quits with chilling AI warning: It could ‘kill us all’ →
14 sept. 2026 · Publication · X
I think it’s possible that the AI companies might, in the next few years, succeed in building superintelligent AI systems that far exceed human capabilities in every domain … Alignment is the problem of preventing this, and is both difficult and unsolved. Our present understanding of how to train AI systems that deeply want what we want is extremely rudimentary.
Program Lead, BlueDot Impact à l’époque
From the same post, as quoted by Ariel Zilber of the New York Post on Yahoo. The first sentence ends without punctuation because the page closes it on an attribution comma ("in every domain," he wrote), which is the outlet's and not his; the ellipsis stands where the reporter's own sentence reports him as "not confident" that such systems would behave as humans intend.
Google DeepMind researcher quits with chilling AI warning: It could ‘kill us all’ →
14 sept. 2026 · Publication · X
I recently resigned from Google DeepMind, where I worked on AGI safety and alignment research. At Google, I witnessed AI development first hand. I too am extremely concerned by the default trajectory of this technology. I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome.
Program Lead, BlueDot Impact à l’époque
The opening of the post he published on X and LinkedIn, sourced to the outlet that reproduces it whole rather than to x.com. The day comes from the X dateline Parameter prints under its copy of the post ("— bilal 🔶 (@bilalchughtai_) September 14, 2026"); The Next Web dates only its own report, 15 September, and the New York Post says "Monday", which is relative dating by the outlet and is not imported. He left DeepMind two months before posting.
Bilal Chughtai left DeepMind’s AGI safety team in July and posted his warning this week →