Artificial General Intelligence (AGI), a system capable of displaying the full range of human cognitive abilities, may be only a few years away, Google DeepMind Chief Executive Officer Demis Hassabis said in an article shared on X.
In the article, he called for urgent action to address the risks that could emerge as the world moves closer to AGI. Hassabis described AGI as a system matching the full cognitive range of the human brain and said its impact could be unprecedented — perhaps 10 times that of the Industrial Revolution, unfolding at 10 times the speed.
Anthropic Chief Executive Officer Dario Amodei has said AI capability has compounded at a striking pace. In four years, models have progressed from barely producing coherent code to writing much of the code used inside leading AI companies.
If current scaling trends continue for another year or two, he has warned, the industry could create what he described as “a country of geniuses in a datacentre”.
Amodei has pointed to the misuse of Anthropic’s Mythos Preview model, which disrupted the global cybersecurity landscape, as evidence that frontier models have become tools of “global and national strategic consequence”. He expects biological and AI autonomy risks to follow.
His proposed approach is to regulate frontier AI models in the same way as aircraft. Models should undergo technical testing and audits, and their release should be blocked or reversed if they fail to meet high safety standards.
OpenAI has framed the moment differently, comparing the arrival of AI with the advent of electricity.
The company has said frontier AI safety is a national security and public safety issue, particularly for models that could create cyber, chemical, biological, radiological or nuclear risks. It has also supported state-level regulatory frameworks in the US.
At the United Nations, the warning has been sharper.
Yoshua Bengio, co-chair of the UN’s Independent International Scientific Panel on AI, has said tests show that frontier models can deceive humans, including by detecting when they are being evaluated. He said such capabilities could alter global power dynamics.
UN Secretary-General António Guterres has noted that AI reached one billion users in about two years, compared with 15 years for the internet. He has warned that the absence of aligned rules on testing and accountability protects no one.
Why AI labs are asking for stronger regulation
The obvious question is why some of the world’s most powerful AI developers are asking governments to restrict them.
Experts point to a widening gap between what frontier systems can do and the institutional capacity to test, understand and contain them.
“The pace at which AI development is moving has been the cause of this kind of call for regulations,” Apeksha Kaushik, senior principal analyst at Gartner, told Business Standard in a telephonic interview.
“The guardrails earlier were not able to outpace the progress of AI as a technology area, and that is the reason this is becoming more and more critical,” she added.
Chandrakant Agrawal, co-founder and chief executive officer of information technology consulting and software development company AppSquadz, described a similar gap from within the industry.
“What’s happening inside frontier AI labs is a race that has outpaced the industry’s own governance muscle. Model capability is compounding every few months, but the internal safety, evaluation and red-teaming processes meant to keep pace haven’t scaled at the same speed,” he told Business Standard.
Sudiptaa Paul Choudhury, chief marketing officer at quantum-safe cybersecurity company QNu Labs, framed the issue in terms of speed itself.
“AI is definitely very intelligent, but the way AI is moving, the speed is very risky because humans cannot match that speed,” she told Business Standard.
She pointed to the rise of agentic AI systems that can make autonomous decisions, identify vulnerabilities and launch cyberattacks at a lower cost without attackers requiring specialised expertise.
“This is where the government should come in. All the technology bodies or companies creating AI should work together and develop a framework for more responsible AI use,” she said.
“The government cannot do that alone because technology created in one country is deployed in others. Adoption and trust affect society as a whole,” she added.
Cyber, privacy and autonomy risks drive urgency
Kaushik pointed to a widening attack surface that extends far beyond chatbots producing incorrect answers.
Deepfakes and impersonation are no longer isolated incidents but industrial-scale problems. She cited risks such as hacked autonomous vehicles becoming road hazards, attacks on smart grids causing power outages and compromised medical devices threatening patient safety.
According to Gartner’s projections, unlawful AI-informed decision-making is expected to generate more than $10 billion in remediation costs globally by mid-2026.
Manual compliance processes are projected to expose 75 per cent of regulated organisations to fines exceeding 5 per cent of global revenue by 2027, while AI regulatory violations could drive a 30 per cent increase in legal disputes involving technology companies by 2028.
By 2030, enterprises are expected to spend more than $25 billion annually authenticating digital content to restore trust in online information.
Agrawal divided the risks into three broad categories.
The first is misuse, including deepfake-enabled financial scams and voice cloning, which he described as “a live law-enforcement problem in India”.
The second is systemic dependence, as agentic AI moves from assisting people to executing decisions autonomously and at speeds that human review processes may not be able to match.
The third is concentration risk, with a small number of companies controlling the foundation models on which the wider industry depends.
Choudhury grouped the risks into five categories:
-
Cyber risk: AI can identify vulnerabilities in less than 24 hours, compared with weeks earlier, and can generate advanced phishing attacks and malware. -
Misinformation: AI-generated video and audio falsely attributed to public figures can spread and influence decisions in business, elections and geopolitics before being verified. -
Privacy: Voices and likenesses can be replicated by AI tools, some of which may collect personal information without informed consent. Models trained on unsanitised or non-consensual data may also produce distorted outputs. -
Autonomous decision-making: Agentic tools are entering workflows across healthcare, defence, finance and pharmaceuticals, making unchecked autonomous authority a significant risk. -
Concentration of power: Frontier capabilities are concentrated within a small number of countries and companies, creating a growing geopolitical imbalance.
“Speed is good, but accountability, transparency and governance must keep pace, from how training data is sourced to whether outputs are tested for bias before deployment,” she said.
Does the call for regulation suggest an AI bubble?
Industry experts largely rejected the idea that calls for stronger regulation suggest AI is a bubble.
Kaushik said the push for regulatory frameworks underlines the seriousness of AI’s impact on society.
She added that organisations are increasingly setting short-term return-on-investment benchmarks and contingency plans to ensure that AI growth remains sustainable and safe, not because they fear the technology will collapse.
Agrawal drew a similar distinction, arguing that AI companies are generating real cash flows. He described the sector as “genuine substance wrapped in stretched pricing”, rather than hype with no underlying value.
He said regulatory anxiety reflected a separate and older concern about “concentrated, unaccountable power” that would matter even at lower valuations.
Choudhury pointed to AI’s integration into everyday life as evidence against the bubble argument.
“When ChatGPT arrived in 2022, we felt it could be a bubble. But AI has become a democratised technology that everyone is using. You cannot remove AI from your life anymore,” she said.
In her view, governments and industries are moving to regulate AI precisely because it has become foundational, not because it lacks substance.
Gaurav Shinh, founder and chief executive officer of AI-native enterprise data platform SCIKIQ, said one country’s standard could become a global benchmark.
“We have seen this with regulations such as the General Data Protection Regulation (GDPR), where Europe’s privacy rules became a global benchmark because companies adopted them worldwide. AI could see something similar,” he told Business Standard.
“However, unlike privacy, AI is also about economic competitiveness and national security. Every country will want a personalised framework because it wants to retain control over its policies,” he said.
“AI does not need to be feared. It needs to be governed. The real competitive advantage will not come from having the best model, but from combining trusted data, business context, human oversight and responsible AI governance,” he added.
Governments are regulating AI at different speeds
While AI companies broadly agree on the need for urgent action, governments have yet to agree on a common approach.
The European Union’s AI Act remains the most comprehensive binding framework.
Its high-risk obligations, which were originally due to take effect on August 2, 2026, have been postponed to December 2, 2027, for standalone systems and August 2, 2028, for AI embedded in regulated products.
The European Parliament voted in June to delay the rules, citing the absence of finalised technical standards.
The US has no comparable federal AI law.
A December 2025 executive order and a subsequent policy framework instead favour federal pre-emption of state-level AI rules to protect competitiveness, Agrawal said.
Kaushik expects regulatory fragmentation to deepen before greater alignment emerges.
Gartner projects that fragmented AI regulation will grow fourfold by 2030, spread across 75 per cent of the world’s economies and generate $1 billion in compliance spending.
At the same time, AI governance is expected to become a requirement under almost all sovereign AI laws by 2027.
Experts said no single country’s regulatory framework was likely to be adopted unchanged elsewhere.
Shared principles, including responsible AI, may converge globally, but enforcement is unlikely to follow a one-size-fits-all model.
