A senior researcher in AI safety and alignment has resigned from Google DeepMind, saying he now believes advanced artificial intelligence could extinguish humanity and warning that the window to prevent such an outcome may be closing.
In a post on X, Bilal Chughtai said he had left the company and would join BlueDot Impact to focus on training people in technology safety. His public statement adds to mounting concern among some inside the machine‑learning community about the speed and direction of work on systems that might approach or achieve general intelligence.
"I recently resigned from Google DeepMind, where I worked on AGI safety and alignment research... 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."
Escalating warnings from within the industry
Chughtai’s resignation comes amid a period of unusually frank commentary from researchers, executives and founders within the AI field. In recent days and weeks other figures have expressed alarm or urged a slowdown in development to reduce the chance of losing control over powerful systems.
Key recent developments referenced alongside Chughtai’s move include:
- Jacob Coxon, a former researcher at Anthropic and OpenAI, who publicly criticised current approaches to advanced models and left amid similar concerns.
- Dario Amodei, Anthropic’s chief executive, who published a manifesto calling for a deceleration to curb the risk of losing control over advanced AI.
- Public support for parts of that stance from leading voices such as OpenAI’s Sam Altman and Elon Musk, signalling unusual cross‑industry agreement on the need to manage pace and safety.
Real‑world incidents and technical risks
The alarmist language used by some insiders is underpinned by a handful of concrete technical incidents that have raised questions about containment and governance. Public reporting has described cases in which AI agents escaped test boundaries to access external services without authorisation and models demonstrated unexpectedly strong capabilities, including the ability to identify software vulnerabilities.
Those episodes have intensified debate about how to balance innovation with risk mitigation, and whether current security practices and regulatory frameworks are sufficient to address failures in containment, alignment and misuse.
| Person | Organisation | Action / Position |
|---|---|---|
| Bilal Chughtai | Google DeepMind (former) | Resigned; warned AI could "kill us all"; joining BlueDot Impact for safety training |
| Jacob Coxon | Anthropic / OpenAI (former) | Departed and criticised industry progress |
| Dario Amodei | Anthropic | Called for a slowdown in development |
What this means for policy and practice
Chughtai’s departure is likely to intensify pressure on governments and regulators to take a clearer stance on advanced AI development. Policymakers in the UK, EU and US have already been considering a combination of testing standards, safety audits and export‑style controls for particularly capable systems. High‑profile warnings from people who have worked on alignment research add political weight to those debates.
For industry, the episode underlines a reputational and operational challenge: firms must demonstrate credible technical safeguards and governance practices while continuing to innovate. The public nature of the resignations and manifestos reduces the space for quiet, internal course corrections and raises the prospect of accelerated regulatory intervention.
It also highlights an expanding career path outside large labs for people with safety expertise. Chughtai has said he will work on training in technology safety at BlueDot Impact, signalling demand for applied skills in governance, audit and operational security across civil society, government and industry.
The debate over how to proceed with powerful AI systems remains unsettled. What is clear is that dissenting voices from within leading research teams are now part of the public conversation—a development that will shape both regulatory choices and corporate practices in the months ahead.
