When AI Researchers Start Quitting: A Warning We Cannot Ignore

On September 15, 2026, Bilal Chughtai — a safety researcher at Google DeepMind who worked on AGI alignment — resigned and published a public statement warning that artificial intelligence "has the potential to kill us all." His departure follows a pattern of exits from AI safety roles at top laboratories, raising a structural question: when the people closest to the technology start walking away, what does the record show about why?

The Resignation That Reverberated

Bilal Chughtai's exit from Google DeepMind was not a quiet departure. He announced his resignation in a post on X and LinkedIn that described working on AGI safety and alignment research at one of the world's most influential artificial intelligence laboratories. His message was unambiguous: humanity may be running out of time to prevent AI from causing widespread, potentially irreversible harm.

"AI has the potential to kill us all," Chughtai wrote, according to multiple reports citing his public statement. He added that things are "about to get crazy" — language that, stripped of its informality, describes a researcher who believes the trajectory of AI development has entered a phase he is no longer willing to be associated with from inside the institution.

Chughtai's resignation did not arrive in isolation.

The previous week, Jacob Coxon, a researcher at Anthropic — another leading AI safety-focused laboratory — also resigned, citing in part his belief that "the people building AI earnestly believe that it could kill everyone, and are building it anyway," according to the Economic Times, which cited Reuters reporting. Two resignations from two of the world's most prominent AI laboratories in less than two weeks constitute a signal worth examining carefully.

Who Is Bilal Chughtai — and What Did He Actually Work On?

Bilal Chughtai worked specifically on AGI safety and alignment research at Google DeepMind. This distinction matters. Not all AI research is safety research. Alignment research — the field dedicated to ensuring that increasingly capable AI systems remain aligned with human intentions and values — is the subfield most directly concerned with catastrophic risk scenarios.

A researcher working in alignment is, by definition, someone whose professional mandate is to identify, model, and mitigate the ways AI systems could behave in unintended or harmful ways as they become more capable. When such a researcher concludes that the environment is moving faster than the safeguards, and chooses public resignation over continued internal advocacy, the institutional implication is significant.

The FreePressJournal reported that two Google DeepMind researchers resigned in connection with this wave of alarm — suggesting Chughtai was not alone within DeepMind itself.

Is This a Pattern — Or Isolated Incidents?

The evidence points toward a pattern, not a coincidence.

Consider the record: Geoffrey Hinton, who won the Nobel Prize in Physics and is widely considered a founding figure of modern deep learning, resigned from Google in May 2023 specifically to speak freely about AI risks. Yoshua Bengio, another Nobel laureate and AI pioneer, has repeatedly warned that AI development is proceeding faster than society's ability to govern it. Sam Altman, chief executive of OpenAI, has himself said in congressional testimony that he is "a little bit scared" of the technology his organisation is building.

In 2024 and 2025, a wave of departures from OpenAI's safety team — including co-founder Ilya Sutskever and safety lead Jan Leike — generated widespread coverage. Leike published a detailed exit statement saying that "safety culture and processes have taken a back seat to shiny products" at OpenAI.

The Squirrels notes that this record does not support a reading of Chughtai's resignation as an anomaly. It is the latest data point in a trend line that stretches back at least three years and crosses multiple institutions.

What Are These Researchers Actually Warning About?

The warnings can be categorised into three distinct risk frameworks, each with a different time horizon and mechanism:

Framework 1: Misalignment at Scale As AI systems become more capable, ensuring they pursue goals aligned with human welfare becomes mathematically harder, not easier. A system optimised for a proxy goal — maximising engagement, maximising efficiency, maximising a defined output metric — may pursue that goal in ways that are technically correct but catastrophically harmful. This is the core concern of alignment research.

Framework 2: Misuse by Bad Actors Capable AI systems — particularly in biological research, autonomous systems, and information warfare — provide leverage to actors who would use them to cause harm at scale. A system that can dramatically accelerate drug discovery can, under different instructions, dramatically accelerate the design of pathogens. This risk does not require AI to "go rogue" — it requires only that dangerous tools become widely accessible.

Framework 3: Institutional Race Dynamics The competitive pressure between leading AI laboratories — and between nation-states investing in AI — creates structural incentives to reduce safety timelines. When the market rewards speed and capability over caution, and when geopolitical competition creates fear of falling behind, safety investment becomes a competitive disadvantage. The resignations from safety roles are, in part, a response to this structural pressure.

Why Are Researchers Choosing Public Exit Over Internal Reform?

This is perhaps the most structurally important question raised by Chughtai's departure.

AI safety researchers at organisations like Google DeepMind and Anthropic are not external critics. They are insiders with access to internal processes, leadership, and technical roadmaps. The fact that they are choosing public resignation — rather than continuing to advocate for change from within — suggests a conclusion that internal advocacy has reached its limits.

Jan Leike's 2024 exit statement from OpenAI made this explicit: he described a situation where safety teams were consistently outpaced by product teams, where safety concerns were raised and acknowledged but did not change the timeline of releases.

The institutional dynamic that emerges from these accounts is one where safety is treated as a department to be consulted rather than a constraint to be respected. Researchers who signed on to that work with a different understanding are, when that understanding proves incorrect, left with limited options.

What Does the Indian Policy Context Mean for This Debate?

For Indian readers and policymakers, the Chughtai resignation is not a distant Silicon Valley story. India's relationship with AI development has several dimensions that make this debate directly relevant.

India is the world's largest consumer of AI-generated content and one of the fastest-growing markets for AI-powered tools in healthcare, agriculture, financial services, and governance. The Indian government has published an AI Mission framework and is actively deploying AI in public services — from predictive policing tools to agricultural advisory systems to automated document processing in the judiciary.

The regulatory framework governing these deployments remains at an early stage. India does not yet have comprehensive AI legislation equivalent to the European Union's AI Act, which came into force in 2024. The governance gap between India's pace of AI adoption and its regulatory infrastructure is, by the metric established by the researchers who are resigning from safety roles, precisely the kind of structural risk they are warning about.

This is not an argument against AI adoption. It is an argument for the sequencing of governance alongside capability — the same argument being made, more urgently, by the researchers walking out of the world's leading laboratories.

What Should Be Done?

The record of what safety researchers have recommended — across public statements, congressional testimony, and published research — converges on several concrete institutional requirements:

1. Independent safety audits before deployment: AI systems above a defined capability threshold should be subject to mandatory third-party safety evaluation before commercial or public-sector deployment. The EU AI Act establishes a version of this framework; India's AI governance architecture should incorporate an equivalent.

1. Structured whistleblower protections for AI safety researchers: Chughtai and his peers are choosing public resignation because internal channels have proven insufficient. Regulatory frameworks should create formal, protected mechanisms for safety researchers to report concerns to independent bodies without career consequence.

1. International coordination on AGI development timelines: No single country or laboratory can unilaterally slow the competitive dynamics that create safety shortcuts. India — as a major AI consumer, a growing AI developer, and a democracy with a stake in global governance — has both standing and interest in participating in multilateral AI safety frameworks.

1. Transparent capability benchmarking: When AI laboratories make claims about the safety of their systems, those claims should be testable against public benchmarks established by independent bodies — not proprietary internal evaluations.

None of these recommendations require accepting the most extreme risk scenarios articulated by departing researchers. They require only accepting that the people who know these systems best have identified a governance gap serious enough to resign over.

FAQ: Bilal Chughtai, Google DeepMind, and AI Safety

Who is Bilal Chughtai and why did he resign from Google DeepMind?

Bilal Chughtai was a researcher at Google DeepMind working on AGI safety and alignment — the subfield concerned with ensuring AI systems behave in ways aligned with human values. He resigned in September 2026, publishing a public statement warning that AI "has the potential to kill us all" and that humanity may be running out of time to prevent widespread harm.

What is AGI alignment research and why does it matter?

Alignment research addresses a core technical challenge: as AI systems become more capable, ensuring that they pursue goals consistent with human welfare becomes increasingly difficult. A system optimised for a measurable proxy goal may achieve that goal in ways that are technically correct but harmful. Alignment researchers work to identify and close these gaps before capable systems are deployed at scale.

Is Bilal Chughtai's resignation unusual, or part of a larger trend?

Chughtai's departure is part of a documented pattern. Within a week of his resignation, Anthropic researcher Jacob Coxon also resigned with similar concerns. Earlier high-profile exits include Geoffrey Hinton from Google (2023) and Jan Leike from OpenAI (2024). The pattern spans multiple leading institutions and multiple years.

What is Google DeepMind's response to these safety concerns?

As of publication, Google DeepMind had not issued a specific public statement responding to Chughtai's resignation post. Google DeepMind has previously published research on AI safety and alignment and maintains a stated commitment to responsible AI development.

What does this mean for India specifically?

India is a major AI adopter across public and private sectors but does not yet have comprehensive AI legislation equivalent to the EU AI Act. The governance gap between adoption pace and regulatory infrastructure is directly relevant to the risks identified by departing safety researchers. India has both standing and interest in multilateral AI safety frameworks.

Conclusion

Bilal Chughtai's resignation from Google DeepMind is one data point. The pattern it belongs to — departures from safety roles at multiple leading AI laboratories, accompanied by increasingly urgent public warnings — is a structural signal that deserves serious analytical attention, not dismissal as alarmism.

The researchers issuing these warnings are not outsiders speculating about technology they do not understand. They are insiders who have seen the systems, participated in the safety evaluations, and concluded that the gap between capability development and safety governance is widening rather than narrowing.

The record does not require accepting the most catastrophic scenarios to justify action. It requires only recognising that when the people closest to a risk start walking away from it, the question worth asking is not whether they are being dramatic — but whether the institutions they are leaving behind are listening.

The data suggests they are not. The question now is whether everyone else will.

Subscribe to The Squirrels newsletter for weekly data-driven analysis on technology, policy, and governance.