Should society dismiss warnings that the technology of artificial intelligence is beyond humans’ control when coming from the scientists who develop the most powerful AI systems? The question becomes especially relevant following a resignation of Jacob Coxon – a former Anthropic researcher who claims that leading AI laboratories “are racing straight to self-improving superintelligence and gambling with our lives.” Indeed, Coxon worked on AI pre-training research at OpenAI and Anthropic.
The argument was further strengthened by the official support provided by Anthropic alignment science lead Evan Hubinger who personally estimates a possibility of mass extinction caused by AI within the next decade at above 10%. This is not a prediction that a disaster is likely to happen but a probability estimate based on personal opinion. But the catastrophic outcome of any such events makes even this small probability worthy of consideration.
There is, however, a trap in viewing this debate in terms of “AI will kill everyone” vs “AI doomers are scaring everyone”. Neither extreme is backed up by scientific evidence. The International AI Safety Report 2026 prepared by the Turing Award recipient Yoshua Bengio with over 100 experts participating in the report concludes that current AI systems lack the capability for causing a catastrophic loss of control. At the same time, there are signs of increasing risk, namely greater autonomy, situational awareness, reward hacking and potentially uncontrollable behaviour observed during experiments. Importantly, expert estimates of future loss-of-control risk vary enormously.
It is also not correct to claim that companies developing AI have no safety plans. Anthropic maintains its Responsible Scaling Policy which establishes that increasing capabilities come along with growing measures of control, risk reporting and third-party reviews. In turn, OpenAI has its Preparedness Framework which defines High and Critical capability thresholds and prescribes necessary safeguards to reduce the risk sufficiently. OpenAI’s 2026 Frontier Governance Framework even includes loss of control among such risks as cyber, chemical, biological and manipulative risks.
What is correct criticism, though, is that we have safety frameworks but we do not yet have a scientifically proven method that guarantees the control of hypothetical superintelligence. AI alignment is an unresolved problem, and the new science of AI control is still very nascent, states the International AI Safety Report. This difference is crucial.
The advantages of conducting frontier-AI research, though, are also very hard to overlook. AI can speed up drug discovery, scientific research, education, engineering, and productivity. AI would be especially useful for India, compensating for the shortage of specialists in healthcare, agriculture, education, and public administration. Therefore, stopping the development of AI could cost society tremendous human and economic losses. Anthropic claims that frontier AI could accelerate scientific discoveries, change healthcare and education and open ways for innovations.
The problem, however, is that competition drives the process. If companies believe that whoever develops transformative AI first gets enormous commercial and geopolitical power, then every company has an interest in accelerating their efforts even if together they would prefer taking a more cautious approach. Competition is explicitly mentioned in the International AI Safety Report as one of the factors making this trade-off more acute.
Therefore, India should reject both technological fatalism and technological complacency. MeitY published India AI Governance Guidelines in November 2025 and later established the AI Governance and Economic Group in April 2026. The opportunity India has is to promote innovation while keeping the supervision: independent evaluations, reporting serious incidents, compute and capability thresholds, protection of whistle-blowers, international cooperation and progressively stronger safeguards as AI becomes more autonomous.
So, is the superintelligence warning an attempt to scare people? Indeed, fear can attract attention and investment as well as political influence. But Coxon’s resignation cannot be considered as a proof of inevitable catastrophe, just as the presence of corporate safety policies cannot serve as such proof.
The most rational position is somewhere in between.
We do not know if superintelligent AI will bring a destruction of mankind. We know that capabilities advance quickly, that there is a real disagreement about the extent of the catastrophe’s danger, and that reliable control of hypothetical superintelligence is not yet developed. The International AI Safety Report expresses this dilemma especially well: current AI systems do not threaten loss of control now but “the decisions made today will determine whether future systems do.”
This makes AI safety not an argument against AI development nor just an attempt at marketing. It is an argument to develop an extremely powerful technology with adequate safeguards.
Humanity does not have to choose between innovation and safety. The wise approach is more challenging: we have to innovate quickly enough to benefit from the extraordinary potential of AI but govern adequately enough to keep intelligence a tool of civilization and not the experiment on it.
Senior Professor and former Head,
Department of ENT-Head & Neck Surgery, Skull Base Surgery, Cochlear Implant Surgery.
Basaveshwara Medical College & Hospital, Chitradurga, Karnataka, India.
My Vision: I don’t want to be a genius. I want to be a person with a bundle of experience.
My Mission: Help others achieve their life’s objectives in my presence or absence!
My Values: Creating value for others.
References:
- Coxon’s resignation and his criticism of Anthropic and OpenAI are covered by the Financial Times, TechCrunchand Fortune.
- Bengio Y, Clare S, Prunkl C, et al. International AI Safety Report 2026. The report synthesises evidence from more than 100 experts and explicitly examines catastrophic loss-of-control scenarios and the considerable scientific uncertainty surrounding them.
- Anthropic. Responsible Scaling Policy, updated August 2026. It describes capability-triggered risk governance, risk reports and safeguards for increasingly powerful frontier models.
- OpenAI. Our Updated Preparedness Framework. April 15, 2025. It defines High and Critical capability thresholds and associated safeguard requirements.
- OpenAI. Frontier Governance Framework. May 28, 2026. It addresses frontier risks including cyber offence, CBRN threats, harmful manipulation and loss of control.
- Government of India/MeitY–IndiaAI. India AI Governance Guidelines were unveiled in November 2025; MeitY subsequently constituted an AI Governance and Economic Group in April 2026.
Leave a reply