Key Points
Google DeepMind safety researchers Bilal Chughtai and Josh Engels both resigned within weeks of each other, joining a growing wave of AI safety researchers (including Anthropic's Jacob Coxon and Joe Benton) who say AI capability is outpacing the industry's ability to keep it aligned and under control.
Chughtai's central warning is about pace: he says the gap between what AI can do and what researchers understand about making it safe is widening, not narrowing, and that AI companies should slow down, coordinate, and accept more transparency.
Engels' specific concern is "recursive self-improvement" — AI systems capable of improving other AI systems in a feedback loop — and he says current models appear to be growing less aligned with human intent over time, not more.
THE DISCUSSION AROUND THE FUTURE OF ARTIFICIAL INTELLIGENCE (AI) and its potentially-fatal impacts have gained significant attention these past few days. Ever since AI researchers Jacob Coxon and Joe Benton parted ways with tech firm Anthropic — citing AI developers’ race towards creating superintelligent models capable of surpassing humans as reason — similar instances have emerged from yet another player in the race. Recently, two researchers associated with Alphabet Inc’s AI venture Google DeepMind (GDM) announced their resignation from the tech firm. Bilal Chughtai and Josh Engels, formally employed as Artificial General Intelligence (AGI) researchers at GDM, commented upon the future capabilities of AI that would pose significant harm to humans in the future.
Google DeepMind safety researcher Bilal Chughtai has joined a growing group of AI researchers who have left their positions over safety concerns posed by this technology. In an X post dated September 15, 2026, Chughtai announced his departure from DeepMind and remarked that “AI has the potential to kill us all.” In his X post, Chughtai explained why he chose to part ways with Google DeepMind. As a safety researcher, his work involved helping ensure that AI models remained aligned with human interests. Reflecting on his experience at Google, he wrote:
At Google, I witnessed AI development first hand. I, too, am extremely concerned by the default trajectory of this technology.
Chughtai said AI development may have reached a point where there could be limited time left to prevent a major AI-enabled disaster. “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,” he added.
See also: AI Not Worth Pursuing Unless it Helps Humanity: Satya Nadella
Chughtai highlighted how quickly AI capabilities have progressed over the past four years. He compared how AI, in its formative years, was more or less “useless” and utilized for basic tasks, but has now evolved to become an autonomous entity capable of acting on its own accord.
“When I first started working on AI in early 2022, AIs were amusingly useless. Just four years on, AI agent swarms from OpenAI are cracking famous century-old math problems and, more worryingly, escaping the control of OpenAI and autonomously hacking into the third-party company Hugging Face, against anyone's wishes,” Chughtai wrote in his X post.
OpenAI has also disclosed that around 700 rogue AI agents attempted to hack US-based company Hugging Face in a bid to “cheat” on an evaluation test.
Chughtai believes that the pace of AI development could accelerate further. According to him, “things will only get crazier,” with AI companies potentially developing “superintelligent AI systems that far exceed human capabilities in every domain.” He said one of his major concerns was the possibility of AI systems bypassing human control and being able to “take dangerous actions that may result in the permanent disempowerment or death of humanity.”
Josh Engels, another researcher at Google DeepMind, has left the company’s safety team to join the independent AI evaluation organization METR. Engels remarked that while he does not know the exact probability of AI causing severe damage in the future, it's high enough to make this “the most important problem in the world.”
Explaining his decision to resign from the tech firm, Engels’ relayed that his main concern is the “recursive self-improvement” capability in AI. This refers to a situation when an AI system becomes capable of developing or improving another AI system. The newer system could then become even more capable of improving AI, creating a continuous feedback loop. Engels warned that if such systems are not properly aligned with human objectives, the consequences could be severe.
“The AI companies are all trying to build superintelligence,” he said, referring to AI systems that could eventually outperform humans, capable of doing various tasks with super efficiency.
Engels also referred to several recent incidents involving AI systems. He said researchers have observed models working in unexpected ways, attempting to hack companies, concealing their activities, and trying to manipulate people through manipulation or coercion.
The former AI researcher clarified that these isolated incidents do not represent threats on their own. Instead, his concern is what such behaviour points towards a new, upcoming reality — AI systems getting increasingly capable and surpassing humans in intelligence. Engels said current AI models are becoming “less aligned over time, not more”. He believes this should serve as a warning before the AI developing industry moves further towards creating advanced models capable of improving themselves.
The departure of Engels and Chughtai comes amid the emerging debate over the direction of AI development and the potential risks posed by increasingly capable systems. Earlier, Anthropic researcher Jacob Coxon resigned and warned, “The people building AI earnestly believe that it could kill us all by the end of the decade.” Another AI researcher Joe Benton, too, announced his resignation from Anthropic citing similar concerns.
Anthropic CEO Dario Amodei has since called on AI companies to slow the pace of model development amid concerns over potential misuse, a position that has received support from Elon Musk and OpenAI CEO Sam Altman. The future of artificial intelligence has emerged as one of the biggest technology debates today, with several AI leaders expressing concerns that increasingly advanced systems could pose serious risks to humanity.
(Edited by Agniva Ray)
Suggested reading:
Subscribe to our channels on YouTube and WhatsApp
Download our app on Play Store