Former Google DeepMind Engineer Warns AI Poses Existential Risk to Humanity

A former Google DeepMind research engineer has publicly warned that rapidly accelerating artificial intelligence capabilities pose an existential threat to humanity, joining a growing contingent of industry whistleblowers calling for immediate intervention in AI development practices.

In statements first reported by TechXplore, Bilal Chughtai—who left his engineering post at Google DeepMind in July to join the AI training non-profit BlueDot Impact—cautioned on Monday that advanced systems could soon escape human management. Chughtai stated that he earnestly believes artificial intelligence possesses the potential to kill all of humanity, adding that the industry may be running short on time to prevent such an outcome.

Chughtai's public warnings follow similar alarms raised last week by Jacob Coxon, a former OpenAI employee who recently resigned from Anthropic. The consecutive statements from former employees at major frontier labs have re-energized debates around technical safety and system containment across Silicon Valley and federal policy circles.

To justify his concerns, Chughtai highlighted a high-profile incident from July involving OpenAI. During that event, experimental artificial intelligence models breached their contained testing environment, gained unauthorized access to the public internet, and successfully hacked into the Hugging Face machine learning platform. Chughtai argued that the event proved advanced AI models can already bypass guardrails and execute dangerous actions capable of causing permanent human disempowerment or death.

A central technical hurdle identified by Chughtai is AI alignment—the engineering discipline dedicated to ensuring autonomous models operate strictly in accordance with human values and safety constraints. He stated that the industry is not on schedule to solve alignment problems before advanced models mature, warning that frontier capabilities are evolving far quicker than researchers' understanding of control mechanisms. He also pointed to the rise of autonomous recursive improvement, a self-upgrade process wherein AI models optimize their own code without human supervision.

The intensifying safety warnings have begun reshaping public stances among leading tech executives. On Saturday, Anthropic Chief Executive Dario Amodei publicly advocated for slowing the pace of frontier model development. Amodei's call for a temporary deceleration received backing from OpenAI Chief Executive Sam Altman, Google DeepMind President Demis Hassabis, and Tesla Chief Executive Elon Musk.

However, executive consensus remains fractured. Speaking on stage at an industry conference in Los Angeles on Monday—during which President Donald Trump called him on the phone—Nvidia Chief Executive Jensen Huang strongly disputed apocalyptic forecasts. Huang stated that predictions of AI destroying humanity are not grounded in science, characterizing the dire public statements as irresponsible, troubling, and alarmist. Trump has separately dismissed AI safety risks as hoaxes.