AI development will not slow down. Governance must evolve as systems move from laboratory research to real-world deployment.

As global leaders convene in New Delhi for the India AI Impact Summit, artificial intelligence (AI) has entered a new phase. The debate is no longer about whether AI will reshape economies, labor markets, and national power. It already is. The more urgent question is whether governments can shape how this transformation unfolds, or whether AI capability will outrun governance altogether.

For the United States and India, which are accelerating cooperation on AI, this matters. AI development cannot be stopped because the economic, military, and geopolitical incentives are too strong to do so. However, it can still be steered if states act before increasingly autonomous systems exceed their ability to govern them.

How Fast AI Is Advancing

The empirical evidence for rapid AI progress is now well established. The Stanford Human-Centered AI Institute’s AI Index 2025 documents several inflection points. Global private investment in AI reached record levels, with more than $100 billion invested in the United States. Performance gains accelerated across reasoning, coding, and multimodal benchmarks. The number of documented AI incidents also rose by 56 percent year over year. At the same time, Chinese software models continued to narrow the quality gap with US frontier systems, particularly in open and semi-open configurations.

Further, AI systems are no longer limited to producing text or images. They can plan multi-step actions, use external tools, write and execute code, and coordinate with other systems. Researchers and industry leaders increasingly describe 2025 as the year agentic AI left the lab, and 2026 as the year of broader agentic deployment.

The Stages of AI Capability and Where We Are Now

Policy debates often group different stages of AI into a single category called “AI.” This often hides capabilities and risk. Academic literature distinguishes between artificial narrow intelligence, systems designed for specific tasks; artificial general intelligence, systems capable of transferring knowledge and solving novel problems across domains; and superintelligence, where machine capabilities exceed human performance across cognitive tasks. 

For purposes of this discussion, artificial general intelligence (AGI) should be understood in a practical and policy-relevant sense, not as a metaphysical endpoint. AGI here refers to AI systems whose general reasoning, memory, and planning capabilities allow them to operate as well as or better than human professionals across multiple domains with significant economic, security, and cultural impact, and to do so without continuous human direction.

Furthermore, within this broader classification, researchers have identified a distinct and operationally important shift toward agentic AI. Agentic AI systems are characterized by their ability to set sub-goals, plan, select tools, and execute multi-step actions over time with limited human supervision, rather than merely producing outputs in response to prompts. 

As of 2025-2026, many advanced AI systems have moved beyond narrow task execution and into this agentic phase, even though they remain short of full artificial general intelligence.

AGI Is Now Inside Policy Planning Horizon

There is no consensus on when AGI will arrive. But experts place it within timeframes relevant to today’s policy decisions.

Metaculus, a forecasting platform, estimates for a weak or nascent version of general AI now range mostly in the late 2020s to early 2030s, with community medians around 2027–2030 depending on the exact definition.

Leaders inside major AI labs have reinforced these estimates. Sam Altman has said that AGI will probably occur during President Donald Trump’s current term, noting that the public underestimates its significance and that breakthroughs could occur sooner than many expect. DeepMind CEO Demis Hassabis has stated that AGI could plausibly emerge within five to 10 years, while Anthropic CEO Dario Amodie believes we will see some form of AGI in 2026, with other predictions suggesting timelines around 2030. 

But the larger point is that the case for early governance does not depend on AGI arriving on any specific timeline. Even if AGI emerges later than expected, the current transition to agentic systems alone is sufficient to justify action, given their labor, security, and governance implications.

AI and Jobs: From Muscle to Mind

The most immediate risk posed by advanced AI is not superintelligence, but labor disruption at scale. As Geoffrey Hinton, the godfather of AI, has argued, the Industrial Revolution made human physical strength largely irrelevant to economic production. AI threatens to do the same to human intelligence. As machines increasingly match or exceed people at routine reasoning, analysis, and coordination, large segments of cognitive labor risk becoming economically redundant. 

Economists have already begun to quantify this exposure. A Goldman Sachs analysis estimates that generative AI could affect 300 million full-time jobs globally and that close to two-thirds of US jobs are exposed to some risk of automation by AI.

This risk is especially acute for India. The country’s IT and business services sector employs millions and underpins export earnings, middle-class growth, and political stability. Agentic AI systems capable of handling end-to-end workflows threaten not just individual jobs but the labor arbitrage model itself, where India houses 16 percent of the global AI workforce. 

What Serious Experts Are Warning About

Some of the strongest warnings about advanced AI come from figures who helped shape it. Eric Schmidt, former CEO of Google, has framed AI less as an apocalypse than as a problem of statecraft. Writing in TIME and Foreign Affairs, Schmidt has argued that advanced AI should be treated like nuclear technology, with state-level oversight justified by low-probability but high-impact risks. Elon Musk continues to frame AI risk probabilistically, estimating a 10-20 percentchance of catastrophic outcomes and arguing that such odds justify aggressive mitigation.

These leaders disagree on tone and timelines, but their statements suggest that they converge on one core point: the risk is not that AI progress will stop; the risk is that governance will arrive too late.

The Core Policy Move: Treat Frontier AI as a Strategic Asset

The most important step the United States and India can take is a conceptual one. Frontier AI should be treated as a national strategic asset, not as ordinary consumer software. It does not mean nationalizing AI development or freezing innovation. It means acknowledging that once systems cross certain capability thresholds, they begin to resemble nuclear technology, advanced cryptography, or space infrastructure more than smartphone applications.

Treating frontier AI as a strategic asset does not mean suppressing open research or centralizing innovation. Effective governance should explicitly protect open-weight model innovation, which has played a critical role in advancing transparency, competition, and distributed research capacity. Open-weight models, when paired with clear capability thresholds and accountability standards, can strengthen resilience, reduce dependence on a small number of closed systems, and improve competition with centralized or authoritarian AI development models.

What the two countries should do is jointly define frontier capability thresholds that trigger heightened scrutiny, require rigorous safety testing and red teaming for systems that exceed agreed reasoning and agentic benchmarks, and limit deployment in especially sensitive domains. It also means treating compute, data centers, and advanced chips as strategic infrastructure, not just commercial inputs optimized for speed and scale.

Why US–India Cooperation Matters

The United States brings frontier research, capital, and compute. India brings scale, talent, real-world deployment environments, and growing influence over AI norms in the Global South. India’s AI Impact Summit in February 2026 positions New Delhi as a credible convening power on safe and trusted AI. When aligned with the United States, the partnership has the potential to shape global norms before fragmented or authoritarian models set them by default.

We cannot and should not try to stop AI development. That would be neither realistic nor desirable, especially in an environment where our adversaries will not slow down. The choice facing policymakers is narrower and more urgent: whether to govern powerful systems early, while leverage still exists, or to react after control has already slipped. 

The evidence from capability trends, labor disruption, expert warnings, and geopolitics all point to the fact that the time for meaningful steering is before superintelligence is fully realized, not after.

About the Author: Kriti Upadhyaya

Kriti Upadhyaya is a visiting fellow for India Policy in the Asian Studies Center at The Heritage Foundation. She currently serves as the vice president of Strategic Advisory at C2Ci Americas, an intelligent platforms company delivering defense and Industry 4.0 solutions globally. Upadhyaya is also the founder of the IndUS Tech Council, a Washington, DC-based policy advisory firm dedicated to strengthening US-India defense and technology collaboration. Previously, Upadhyaya was an associate fellow for the Wadhwani Chair in US-India Policy Studies at the Center for Strategic and International Studies (CSIS). She has held various policy and business development roles in India and the United States in the past nine years, including a tenure at Adani Defense and Aerospace.

The post AI: Why We Can’t Stop (But Must Steer) appeared first on The National Interest.