Google DeepMind Researchers Quit As AI Safety Warnings Grow Over Superintelligence Risks

Former Google DeepMind researchers are raising increasingly stark warnings about the pace of artificial intelligence development, with Josh Engels leaving its AGI safety team and Bilal Chughtai separately warning that advanced AI could pose catastrophic risks to humanity.

 

TL;DR

 

• Josh Engels left Google DeepMind's AGI safety team to join independent AI evaluation group METR.
• Engels warned there is a “terrifying chance” AI could cause immense harm within five years.
• Former DeepMind researcher Bilal Chughtai separately warned AI could potentially threaten humanity.
• Their warnings come as researchers and lawmakers push for stronger oversight of increasingly capable AI systems.

 

Josh Engels, who worked on Google DeepMind's artificial general intelligence safety team, revealed that he left the company to join nonprofit AI evaluation organization METR.

According to The Indian Express report on Josh Engels, Engels said he had left DeepMind around three weeks before publicly announcing the move on September 13. He said he enjoyed his work at the company and had also turned down offers from Anthropic and OpenAI.

His reason was the perceived scale of the risk.

Engels said he now believes there is a “terrifying chance” increasingly capable AI systems could cause immense harm within the next five years if safety and alignment work fails to keep pace with improvements in AI capabilities. He argued that “we need more time,” calling for AI development to be paced so that capabilities do not outrun researchers' ability to keep systems aligned.

Having moved to METR, Engels plans to study how AI misalignment emerges during training, whether existing safeguards are sufficient and whether current research is on track to solve alignment problems.

METR, or Model Evaluation and Threat Research, independently evaluates advanced AI systems for potentially dangerous capabilities and broader societal risks.

 

Another Former DeepMind Researcher Raises A More Severe Warning

 

Engels is not the only former Google DeepMind researcher raising concerns.

Bilal Chughtai, a former research engineer who worked on AGI safety and alignment at DeepMind, said he resigned from the company in July 2026.

According to Reuters' September 15 report on Bilal Chughtai, Chughtai said he believes AI has the “potential to kill us all” and warned that time could be running out to prevent such an outcome.

Chughtai argued that developing advanced AI safely remains possible, but would require coordination across the industry to prevent what he called a “manic race between AI companies.”

He called for AI development to move at a pace society can handle, giving researchers and policymakers time to address emerging risks before they result in extreme harm.

Google did not respond to Reuters' request for comment outside regular business hours regarding Chughtai's warning.

The two researchers are making distinct claims. Engels warned about a significant possibility of immense harm within five years, while Chughtai went further by explicitly discussing the possibility of human extinction.

AI Safety Concerns Are Spreading Beyond Google DeepMind

 

The warnings come amid a wider debate inside the AI industry over whether frontier AI development is progressing faster than safety mechanisms and regulation.

Anthropic researcher Jacob Coxon recently resigned while warning that people building advanced AI genuinely believe it could threaten humanity before the end of the decade. Anthropic scientist Evan Hubinger subsequently said he believed there was a greater than 10% probability of AI killing all humans within the next decade.

Engels and former Anthropic safety researcher Joe Benton have both joined METR, which focuses on independently testing whether advanced AI systems could develop capabilities that pose serious risks to society.

The debate is also beginning to influence policymakers.

According to Reuters' report on proposed US AI safety rules, US senators are discussing legislation that could require developers to show they are taking reasonable precautions against potentially catastrophic AI harms. Proposals under discussion include government testing of AI systems and mechanisms for restricting models considered unsafe.

However, there remains significant disagreement over how quickly AI should advance and how heavily it should be regulated.

For Engels and Chughtai, the concern is that the window for solving difficult alignment and safety problems could narrow as AI systems become more capable.

Their departures add to a growing group of researchers willing to leave leading AI labs and publicly argue that safety efforts are not keeping pace with the race to build more powerful systems.