‘We might be running out of time’: Google researcher warns on AI
A Google researcher has resigned from DeepMind, warning that fast-moving AI development could outpace safety research and raise serious risks for humanity.
Bilal Chughtai, a London-based safety researcher who worked at Google DeepMind, has resigned and raised concerns about how quickly artificial intelligence is advancing.
In a social media post, Chughtai said he had stepped down from Google DeepMind, where his work focused on AGI safety and alignment research. He said he became deeply concerned about AI’s “default trajectory” after seeing its development up close.
His strongest warning was that AI could “kill us all”, and that humanity may be running out of time to prevent such an outcome.
The risk of increasingly capable AI
Bilal's main concern is the possibility of superintelligent AI. These systems could potentially outperform humans across almost every field, including reasoning, planning, scientific research and real-world decision-making.
Chughtai argued that AI progress has accelerated sharply in recent years. Systems that were once limited and unreliable are moving towards more capable AI agents that can plan tasks, use tools and operate with less human input than traditional chatbots.
That brings forward another issue: control.
If such systems are not properly aligned with human goals, they could behave in ways their creators did not intend. AI alignment research focuses on making AI systems follow human instructions safely, rather than misinterpreting goals or taking harmful shortcuts.
Why his resignation matters
Bilal Chughtai’s exit comes amid a wider debate over how quickly AI development should move. The Hindustan Times report noted that his resignation followed similar concerns from Anthropic researcher Jacob Coxon.
It also pointed to public calls from major AI leaders for greater caution as companies compete to build more powerful systems. Chughtai is not arguing that disaster is certain. His concern is that AI capabilities may be advancing faster than researchers’ understanding of how to reliably control them.
That tension sits at the heart of the AI safety debate. More capable systems could bring major benefits to healthcare, education, productivity and scientific research. But safety researchers argue that stronger systems also require stronger safeguards, particularly when they can take actions beyond generating text.
Can AI development be made safer?
Despite the severity of his warning, Bilal ended with cautious optimism. He said safely navigating AI remains possible if companies avoid an uncontrolled race and improve coordination and transparency.
His resignation puts a difficult question back in focus: can AI safety research keep pace with the technology it is meant to make safer? For companies building increasingly capable systems, the answer could shape how far and how quickly AI develops next.