POONAM SHARMA
What if artificial intelligence becomes more powerful than our ability to understand and control it?
The question has become more urgent since AI safety researchers resigned from some of the world’s top artificial intelligence companies. Their resignations, warnings and subsequent work at independent AI-safety organizations have brought the debate about the future of AI to the global forefront again.
One of the more recent developments involves Josh Engels, a researcher on the AGI Safety Team at Google DeepMind. Engels left DeepMind and joined **METR** (Meta-Experimental Technology Research), a non-profit focused on AI safety and investigating AI misalignment incidents.
Misalignment occurs when the objectives or behavior of a system do not correspond to the intentions of humans. His personal website says that his current work at METR is about investigating AI misalignment incidents and mitigating the potentially catastrophic risks associated with increasingly powerful artificial intelligence systems.
His decision leads to an uncomfortable but important question. As artificial intelligence systems become more capable, autonomous and sophisticated, will we still be able to understand why they do what they do? More crucially, will humans always be able to stop them if their behavior goes in an undesirable direction?
The concern is not about chatbots today But it is important not to misunderstand the concerns of AI safety researchers.
Their warnings are not meant to suggest that today’s ChatGPT or other mainstream artificial intelligence systems are on the verge of destroying humanity. To read it that way would make a serious scientific and technological debate into a piece of sensationalism.
But what people really are worried about is future generations of super advanced AI — such systems might be able to work with more autonomy, do increasingly complex tasks and maybe even help improve their own capabilities.
But if those systems become substantially more capable, the balance between human control and machine capability could become increasingly unmanageable.
That’s exactly why AI safety has become a serious field of research within the tech industry itself.
Google DeepMind has shared publicly about efforts including its *AGI Safety Council, Responsible AI work and Frontier Safety Framework, which aim to identify and reduce possible extreme risks from advanced AI.
The message is not, therefore, that AI itself is inherently dangerous. Rather it is that as a technology becomes more powerful it requires more advanced safeguards.
More questions are being asked by more researchers
But Josh Engels is not the only researcher to have left or issued warnings that have fueled this debate.
Another researcher connected to Anthropic, Jacob Coxon, also recently resigned for concerns about AI safety. Coxon said there is a growing “race” among AI companies to produce more powerful and self-improving artificial intelligence systems, and warned such competition could be dangerous.
Evan Hubinger, who works on alignment science at Anthropic, is also in the discussion. He has talked about the possibility of very serious consequences from advanced AI.
According to a report in Scientific American, Hubinger has estimated more than a 10 percent chance that AI will pose an existential threat to humans in the next decade.
That figure, however, is to be understood correctly. This is a personal probability estimate by one researcher, not an established scientific prediction or a certainty about the future.
And that distinction counts.
It is very different to warn of a possibility and to predict an inevitable outcome. Another Warning From a Former DeepMind Researcher
Former AI safety researcher at DeepMind Bilal Chughtai has added another dimension to the debate with public warnings.
On September 15, 2026, Reuters reported that Chughtai had publicly warned of potentially extremely serious consequences associated with AI. He departed DeepMind in July 2026, the report says.
But at the same time, Chughtai has argued, safe AI development is possible, but it will take more coordination and regulation among AI companies.
This point warrants special attention.
The talk is gradually moving away from the simple question of whether AI should exist to the much more practical question:
What rules, safeguards and international mechanisms are needed to keep ever more powerful AI under meaningful human control?
Real-World Dangers of AI Already Exist
You don’t need to wait for some hypothetical superintelligent machine to be able to speak about AI risk.
Some of the risks are already visible.
AI-generated misinformation can spread like wildfire Deepfakes can generate fake audio, video and images that become increasingly difficult to distinguish from reality. Cyber risks can become more sophisticated. Decisions can be influenced by algorithmic bias. An over-reliance on AI could also dull human judgment and critical thinking.
These are not future scenarios. They are part of the technological landscape we already have.
So the potential dangers of future advanced AI are an extension of a broader question: How do societies manage powerful technologies whose capabilities can develop faster than the rules governing them?
Could AI Destroy the World?
There’s no scientific answer that allows us to say that AI will definitely destroy humanity.
Some researchers argue that the growing autonomy and power of AI could pose catastrophic, or even existential, risks. Other are far more wary of such worst-case scenarios, and question both their probability and the assumptions on which they are based.
We need to see both the possibilities and the uncertainties.
The responsible attitude is not to disregard the warnings, nor to proclaim the apocalypse as inevitable.
Fear is not evidence.
At the same time uncertainty should not be an excuse for ignoring potentially serious risks .
AI Power Race Requires Guardrails
Perhaps the biggest concern is not AI per se, but the race to build ever more powerful AI systems.
Commercial and strategic pressures can drive rapid development as companies compete to build faster, more capable and more autonomous systems.
This makes independent safety testing all the more important.
“AI companies need strong internal safety mechanisms, but external auditing provides an additional layer of review. Regulation can set minimum standards. International cooperation can tackle risks that no single country or company can face on its own.
The aim may not be to put a stop to technological progress.
It should be ensuring that technological progress does not outpace humanity’s ability to cope with its consequences.
The exchange between Sam Altman of OpenAI, Dario Amodei of Anthropic and Demis Hassabis of Google DeepMind reveals just how central AI safety and regulation have become to the wider AI debate.
But this is a debate that needs to include governments, scientists, tech companies and civil society.
We need processes for independent safety assessments, greater accountability, transparent risk reporting and meaningful international cooperation.
Most importantly, humanity must be able to intervene if an AI system is acting in ways that represent unacceptable risks.
The Real Sign of Trouble The resignations and warnings of Josh Engels, Jacob Coxon and Bilal Chughtai are not proof that an AI disaster is unavoidable.
But they should not be dismissed out of hand either.
Their concerns underscore one of the core dilemmas of the technology age: we are building machines of ever more extraordinary powers, and we are still learning how to govern those powers responsibly.
Artificial intelligence can revolutionize medicine, education, science, governance, industry and everyday life. The potential upside is massive.
But power without adequate checks and balances can pose risks.
So, the goal should not be to run away from AI because of fear. And mankind should not blindly embrace every new capability just because it is technologically feasible.
Safe, responsible and accountable AI development must be the goal.
And there is one principle that is absolutely non-negotiable:
Humans need to maintain meaningful control over artificial intelligence.
But the consequences could be very hard to manage if we ever build systems with capabilities that outstrip our ability to understand, oversee or stop them.
So the future of AI is not technological.
It’s a matter of human accountability.
The machines could become super intelligent. But we must keep the responsibility for how that intelligence is created, controlled and used with humans.
World moves toward a system where those affected by global decisions have a meaningful say in them.
