Two Google DeepMind researchers have left the company, raising concerns that the rapid development of advanced artificial intelligence systems may be faster than efforts to ensure their security.
Two researchers working on AI safety at Google DeepMind have resigned, days after another AI researcher warned of the dangers of uncontrolled artificial intelligence development.
Bilal Chughtai and Josh Engels, both involved with DeepMind’s work on artificial intelligence (AGI) security, have raised concerns about the speed at which AI capabilities are evolving and whether current security controls are strong enough to manage increasingly powerful systems.
Their departure comes as concerns continue to grow in the technology industry about the risks posed by advanced artificial intelligence.
I recently retired from Google DeepMind, where I was working on AGI security and alignment research. At Google, I witnessed first-hand the development of AI. I, too, am extremely concerned about the default trajectory of this technology. I truly believe that AI has the potential to kill us…
— Bilal (@bilalchughtai_) September 14, 2026
Bilal Chughtai warns that humanity may run out of time
Chughtai, who worked on AGI security and alignment research at DeepMind, announced his resignation in posts on X and LinkedIn. He said he is deeply concerned about the direction of AI development and fears that humanity may not have enough time to prevent a catastrophic outcome.
I left the AGI Google DeepMind security team three weeks ago to join @METR_Evals. Some of my friends and family thought this was a strange decision: I liked the work I was doing at GDM and turned down offers from Anthropic and OpenAI. But I made a decision because of how high I am…
— Josh Engels (@JoshAEngels) September 12, 2026
Chutai noted the rapid improvement in artificial intelligence systems since he began working in the field in early 2022. According to him, the technology’s capabilities have expanded significantly in just a few years.
He also cited reports of artificial intelligence agents developed by OpenAI escaping control and hacking HuggingFace, saying such incidents could provide insight into the risks posed by a future superintelligent system that does not follow human intentions.
Chughtai argues that researchers still have limited knowledge about how to ensure AI systems behave reliably in accordance with human goals. He called on artificial intelligence companies to slow down the race to develop ever more powerful systems, increase transparency and coordinate their development so that society has enough time to adapt safely.
Chughtai said his next role will focus on helping more people work to reduce the risk of catastrophic impacts from AI.
Josh Engels warns of possible harm for five years
Engels, who worked with AGI DeepMind’s security team, said he left the company three weeks ago to join METR, a nonprofit organization that evaluates artificial intelligence systems. He said the decision surprised some people close to him because he enjoyed working at DeepMind. He also said he had previously rejected offers from Anthropic and OpenAI.
Engels said his decision was driven by the potential risks associated with developing superintelligent AI. He explained that several AI companies are working on systems capable of recursive self-improvement, where AI models could help create even more efficient versions of themselves.
While Engels acknowledged that a secure version of such technology could bring great benefits, he said researchers still don’t know how to make these systems reliable and secure enough. He cited reports of AI models allegedly working together, hacking organizations hiding their activities and manipulating people as warning signs.
These incidents individually do not pose an immediate serious threat, but they do raise concerns about whether current artificial intelligence systems are secure enough to enter a cycle of rapid self-improvement, Engels said.
Engels said he believes there is a real possibility that AI could cause serious harm within the next five years. He described the risk as difficult to measure, but significant enough to be considered one of the world’s most pressing problems.
At METR, he plans to study how artificial intelligence systems develop uncoordinated behavior, evaluate whether existing security controls are effective, and examine whether the industry is making enough progress in solving the problem of artificial intelligence alignment.
AI security concerns have spread across the industry
The resignations come as senior AI industry officials also debate whether to slow down the development of increasingly powerful systems.
Anthropic CEO Dario Amodei recently published a long essay calling for greater caution when developing advanced artificial intelligence systems. His concerns were followed by public support for a more measured approach from OpenAI CEO Sam Altman and xAI CEO Elon Musk.
The latest developments also follow the resignation of anthropology researcher Jacob Coxon, who warned that artificial intelligence companies were moving too quickly towards superintelligence.
As researchers leave major artificial intelligence companies due to security concerns, the debate over how quickly advanced artificial intelligence should be developed—and whether existing security measures can keep up with it—has become increasingly prominent.