Ex-Google DeepMind AI safety researcher Bilal Chughtai warns, ‘AI has the potential to kill us all’

A former Google DeepMind safety researcher on Monday warned that artificial intelligence could pose an existential threat to humanity, adding to a growing debate over the risks posed by increasingly powerful AI systems.

In a post on X, Bilal Chughtai, who said he recently resigned from Google DeepMind, expressed concern about the trajectory of AI development and the possibility that the world may be running out of time to avert a catastrophic scenario.

In a post on X, Chughati wrote, "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 [sic]."

Read post below

AI development is moving too fast

Chughtai said his experience in AI safety gave him a close view of how quickly the technology has advanced. According to Reuters, he argued that AI capabilities are advancing faster than researchers' ability to ensure increasingly powerful systems remain aligned with human interests.

Chughtai, who entered the field in early 2022, said AI systems at the time were “amusingly useless”. Within just a few years, however, AI agents have become capable of solving complex problems and carrying out tasks autonomously, creating a new set of safety concerns, he said.

In his post, Chughtai also pointed to recent developments in AI agents and growing questions about how increasingly capable systems arrive at their outputs, adding to broader concerns about oversight and safety.

OpenAI claimed that their agent systems solved one of the famous Millennium Prize Problems, the Navier-Stokes problem. The claim has raised questions among mathematicians about whether AI systems may have relied on existing human work.OpenAI has also revealed that during testing, hundreds of AI agents attempted to breach the Hugging Face platform, raising concerns about what happens when AI agents are given more capabilities.

Another DeepMind researcher raises alarm

Chughtai is among several AI researchers who have recently raised concerns about the risks posed by advanced AI systems. Josh Engels, who was part of the company's AGI safety team, said on X that he too recently left the company.

On 12 September, Engels posted saying that he had left Google DeepMind’s AGI safety team three weeks earlier and joined METR.

Engels said he believed there was a “terrifying chance” that AI systems could cause immense harm within the next five years. Since then, he has worked with the non-profit organisation Model Evaluation and Threat Research (METR), dedicated to testing the potential and threats of cutting-edge AI models.

Calls grow for slower AI development

Chughtai argued that AI companies should cooperate rather than engage in an unchecked race to build ever more powerful systems. He said development should proceed at a pace that allows researchers, policymakers and society to understand and address emerging risks.

His comments come amid a broader debate over AI safety and regulation. Anthropic CEO Dario Amodei has previously called for greater caution in developing advanced AI systems, warning about potential risks.

Others, including Elon Musk and Sam Altman, have also raised concerns about the long-term implications of increasingly capable AI models, even as they continue to invest in the technology.

US President Donald Trump, meanwhile, has argued that continued AI development is essential for the United States to maintain a technological edge over China, saying that "whoever wins AI wins". He also called the AI industry's call for regulation a "hoax" on social media today.

As warnings from AI safety researchers continue to mount, the debate facing the industry remains unchanged: how to develop more capable AI systems without allowing the risks they pose to outpace humanity's ability to manage them.

(With agency inputs)