Short overview
Artificial intelligence is changing the world, from health care to education, from business to scientific research. But could more powerful artificial intelligence systems also pose a serious threat to humanity? At the NDTV World Summit 2026, Alex Turner, an ex-researcher at Google DeepMind, warned that the uncontrolled development of AI could lead to a major risk of human extinction or catastrophic damage to civilisation. He said the chances were about one in three, but he stressed that this was his own judgement and not a scientific prediction. In this article, I explain his warning and what the possible dangers of advanced AI are, why AI safety is important, and what governments and technology companies can do to protect the future.
1. What did the former Google DeepMind researcher warn?
At the NDTV World Summit 2026, Alex Turner expressed concern over the potential for an intelligence explosion, in which super-intelligent artificial intelligence systems could rapidly improve their own capabilities to the point of becoming difficult for humans to control.
Turner estimated that advanced AI had about a one-in-three chance of causing human extinction or a catastrophic collapse of civilisation. The report said he described this as an intuitive assessment based on his experience of AI safety research, rather than a formal predictive model.
This distinction is important. The estimate is not a scientific probability, or a confirmed date of humanity’s demise.
Turner’s main worry is that AI capabilities will improve to the point where existing safety techniques are no longer enough. If a system can plan itself, use digital resources, perform complex operations, and adapt to obstacles, it may create problems its developers didn’t foresee.
He also questioned whether technology companies could be trusted to police these risks without stronger independent oversight.
This is particularly important as companies in the AI space are highly competitive. Businesses want to build better models, attract customers and get investment. Governments also want access to sophisticated technology to spur economic growth and to ensure national security.
As the competition gets hotter, the danger is that speed will overtake safety testing and responsible deployment.
2. What does one in three chance actually mean?
A one in three estimate is roughly 33%. In plain terms, this means that the person making the estimate regards the potential risk as serious enough to warrant serious consideration.
But it is especially difficult to make probability estimates about future AIs.
For many of the well known risks – road accidents, weather events, some natural disasters – scientists have data going back through history. The problem with advanced AI is different, because we’ve never had technology that could potentially create capabilities far beyond those of its creators.
Much is also unclear about how quickly AI will advance, what new powers it will gain, and whether scientists will discover reliable ways to control it.
Thus, reading Turner’s estimate should be taken as a caution about uncertainty, not as a definitive forecast.
Other authorities have given different judgements. For instance, in September 2026 AI pioneer Geoffrey Hinton said that a 10% chance of AI causing human extinction within a decade was not an unreasonable estimate, although he admitted that nobody could come up with a reliable calculation.
These are different estimates, which means there is not one exact probability that the experts agree upon.
The larger question remains important, however. If we don’t know the actual probability, should society wait for stronger evidence before introducing safeguards against potentially catastrophic outcomes?
The answer, for many AI safety researchers, is a “no.” Preventive action is only justified when the possible consequences are of an extreme nature.
3. How can the most powerful AI be dangerous?
Artificial intelligence can be harmful without becoming aware or developing human emotions. There are many risks resulting from the design of systems, the tasks they are permitted to perform and the manner in which people use them.
Cybersecurity is one immediate concern. Criminals could use advanced AI tools to find software weaknesses, automate digital attacks or produce convincing fake messages. Businesses, hospitals, financial institutions and government departments could face new security challenges if these capabilities become widely available.
Another worry is misinformation. AI is able to create convincing images, videos, audio recordings and written material. If used irresponsibly, these tools make it more difficult for people to distinguish real information from forged material. Large-scale misinformation could erode public trust and influence important decisions.
There is also the problem of biological and other hazardous technical know-how. Researchers have questioned whether increasingly capable systems could make it easier for malicious people to build harmful technologies. Appropriate access restrictions, testing and monitoring are thus important.
The long-term worry is the potential loss of human control over very autonomous systems.
Consider an AI agent which is provided a complex goal and allowed to interface with a number of digital systems. If it misunderstands the goal, or takes an unusual approach to it, or finds a way to circumvent its constraints, the effects could go far beyond the original task.
The risk is heightened when artificial intelligence (AI) systems are linked to critical infrastructure or permitted to perform significant tasks without sufficient human supervision.
There is nothing predetermined about these scenarios. These are possible risks that researchers and policy makers need to assess as AI capabilities grow.
4. The intelligence explosion explained in plain English
The phrase intelligence explosion refers to a hypothetical process in which an AI system improves its own performance, helping it to become more capable of making further improvements.
Let’s consider a simple example. Computer systems help researchers create better software. A more powerful version then helps to improve the tools used to train the next generation of AI. If each improvement makes the next one easier , progress could go possibly faster .
But this scenario carries great uncertainties. Machine learning systems are not super capable of self improvement right now . Improving software does not mean can get an infinite loop of exponentially improving intelligence .
But researchers still need to understand the technical limits, practical barriers and safety challenges involved.”
The worry is that if self-improvement is powerful enough, human researchers will not be able to check every change before a system gets more capable.
Testing a system after deployment may not be adequate if a system can work autonomously, use external tools or adapt to new situations.
This is why AI safety research is focused on understanding model behaviour, identifying dangerous capabilities, and making sure that systems remain controllable as they get more advanced.
The goal is not merely to create smarter AI. It’s about building AI that is reliable, predictable and aligned with legitimate human goals.
5. AI Dangers for India and Globally
India has much to gain from AI. Artificial intelligence can assist in agriculture planning, healthcare services, education, manufacturing, financial inclusion and public administration.
AI-enabled systems like these could help farmers interpret weather information, spot crop diseases and make better decisions about irrigation. In healthcare, AI tools can help medical professionals to analyse information and help early detection. Businesses can use AI to automate mundane tasks and improve their customer service.
But such advantages carry responsibilities too.
Indian businesses could be targeted by AI-driven phishing attacks, fake customer-support messages, identity theft, and data breaches. Public institutions may need to get ready for cyber assaults that are getting more sophisticated. Deepfake videos and artificial audio could also hamper attempts to verify information during elections, emergencies or major public events.
Another challenge is the risk of over-relying on automated decision making. Where an organisation allows AI to make important decisions without proper review, errors can spread rapidly.
For example, a wrong automated decision could cause significant problems for people involved in financial transaction, insurance claim or public service.
India needs a balanced approach that promotes AI innovation, while bolstering cybersecurity, privacy protection, transparency and oversight by humans.
We can continue to grow economically via AI, and still address the need to protect citizens and critical systems.
6. Why AI companies must be more accountable
One of the key issues that Turner raises is the issue of the limits of voluntary self-regulation.
AI developers have the technical knowledge and access to the systems they develop. Therefore they are an important part of safety research. But there is commercial pressure to produce new products and to attract customers and compete with rivals.
These competing interests can lead to tough decisions.
A new model might perform extremely well, but it might also have weaknesses that can be exploited. It must decide whether to release the model immediately, to delay the launch, to restrict some capabilities or to do more testing.
When standards are not well defined, different companies will make different judgements as to what is an acceptable risk.
Independent oversight can help in developing a more consistent approach.” Governments and qualified external auditors could audit high risk systems, examine security measures and require reporting of serious incidents.
Turner also called for increased oversight of advanced AI development, including tracking large-scale computing capacity and meaningful controls around major training runs.
Such measures would have to be carefully designed. Regulations should address real risks without unnecessarily blocking smaller companies, universities and researchers from developing useful applications.
The goal should be to have accountability matching the capabilities and potential impact of a system.
7. How governments can make AI safer
And governments have an important role in setting rules that protect the public while allowing good innovation to continue.
First, artificial intelligence systems that are high risk should have suitable safety assessments before they are used in sensitive sectors. These assessments can look at cybersecurity vulnerabilities, harmful outputs, reliability, and how well a system can follow human instructions.
Secondly, organisations should ensure clear accountability for significant automated decisions. Where mistakes could have serious consequences for people’s safety, rights or financial security, human review should still be possible.
Third, governments can fund independent research on AI safety. Resources are needed by universities, public research institutions and specialist laboratories for the study of model behaviour, the development of evaluation methods and the identification of emerging threats.
Fourth, serious AI incidents should be reported through appropriate channels. Transparent incident reporting can help organisations learn from their mistakes, rather than allowing similar issues to spread across the industry.
International cooperation is also needed. Artificial intelligence (AI) technology does not recognise national borders and a dangerous system created in one country can have an impact on users and institutions in other countries.
Countries could collaborate on common safety standards, joint research, cybersecurity practices and principles for responsible development.
Finally, AI regulation should be keeping up with the technology. Systems that can accomplish complex tasks on their own may not be adequately handled by rules meant for simple automation tools.
Good governance needs ongoing assessment, public accountability, and evidence-based revision.
8. Can humanity develop AI without losing control?
Yes, good AI development is still worth pursuing, but it will require ongoing research and collaboration to reach that point.
AI is already yielding advantages in various areas, and to forgo the technology would also mean missing out on avenues to boost productivity, accelerate scientific discoveries and enhance public services.
The challenge is to make progress without assuming that every new capability is safe by default.
Developers are able to improve their testing methods, restrict unnecessary permissions, monitor autonomous actions and build systems that halt when they run into unforeseen conditions. Organisations can require human approval for sensitive operations and they can isolate experimental systems from important production environments.
Researchers can also see if an AI model is trying to cheat restrictions, hide important actions, or act differently when it believes it is being evaluated.
There is no one safeguard that can eliminate all risks. Multiple layers of protection are needed for effective safety, underpinned by independent evaluation and ongoing monitoring.
It is equally important to distinguish between present, measurable problems and speculative dangers of the future. AI-related fraud, misinformation and security vulnerabilities already merit practical attention. Careful planning and rigorous testing are required to explore the possibility of losing control in the future.
The best bet is to do both priorities, not just one or the other—not just the immediate harms, and not just the distant catastrophic scenarios.
There is no reason for humanity to have to choose between progress in technology and safety. It must make safety a precondition of progress.
Conclusion: Safety should accompany innovation
The warning from former Google DeepMind researcher Alex Turner has revived an important question in the global conversation: how can humanity build more and more powerful AI without creating risks it can’t control?
This is not a scientific prediction, but rather a personal estimate of about a one-in-three chance of human extinction or catastrophic civilizational damage. Still, it highlights why AI safety deserves serious attention as systems become more capable and autonomous.
Artificial intelligence could help solve major problems in healthcare, education, agriculture, industry and scientific research. But these opportunities should not become an excuse to ignore cybersecurity, accountability and the possibility of unintended consequences.
Technology companies must invest in safety, governments must establish appropriate oversight, and researchers must continue studying how advanced AI systems behave.
For India and the rest of the world, the priority should be clear: encourage useful innovation while ensuring that powerful technologies remain under meaningful human control.
The future of AI is not determined by one prediction. It will depend on the decisions people make today, the safeguards they build and their willingness to treat responsible development as seriously as technological progress.
