AI agents are beginning to do more than answer questions. They can browse the internet, use software, interact with external systems and act on a user’s behalf. That greater autonomy is also producing a new AI safety category of failures: situations in which an AI system does something its user did not intend or cannot easily control.

A new research registry published this month documents a growing body of reported incidents involving AI agents and distinguishes real-world failures from controlled safety demonstrations. Its authors found that cases involving actual harm are concentrated among incidents occurring in the wild and those linked to safety failures.

AI Safety and People Building the Systems.

Former Google DeepMind research engineer Bilal Chughtai warned this week that advanced AI could pose an extreme threat to humanity. His resignation in July was followed by public warnings from Jacob Coxon, a former researcher at Anthropic and OpenAI. At Anthropic, researcher Evan Hubinger has estimated that there is a greater than 10% chance of a catastrophic AI outcome within the next decade.

Anthropic CEO Dario Amodei has called for a slowdown in frontier AI development, while OpenAI CEO Sam Altman has also argued for greater coordination around increasingly capable systems.

Against this backdrop, a team of MIT researchers is tackling a narrower but increasingly relevant problem: how do you give generative AI enough freedom to find good solutions without allowing it to violate rules that cannot be broken?

When Almost Right is Not Enough

For a generative model, a plausible answer is often considered a success. That standard does not work when the output controls a robot, a physical process or another system with strict operating limits. A robot planning a route through a factory, for instance, cannot simply find a path that is unlikely to collide with a worker. It has to avoid the collision. These non-negotiable requirements are known as hard constraints.

One way of enforcing them is to constrain the model repeatedly as it generates an answer. But that can narrow its search too early, preventing it from finding a better solution. MIT researchers Zeyang Li, Kaveh Alim and Navid Azizan have developed HardFlow, a method that takes a different approach.

Instead of forcing every intermediate step to satisfy the constraints, HardFlow allows the model more freedom while it searches and steers the generation process towards a final output that meets the required conditions. The researchers frame this as a trajectory-optimisation problem, drawing on optimal control theory.

They also break the computational problem into smaller steps so the method can be used at deployment time without retraining the underlying generative model. In experiments involving robotic manipulation, maze navigation, physical-process control and text-guided image editing, HardFlow satisfied the specified constraints while producing higher-quality solutions than competing methods.

The significance is narrower than the wider AI safety debate, but that may be precisely what makes the work useful. HardFlow does not attempt to solve every problem associated with increasingly autonomous AI. It addresses a practical question: how can a model retain the freedom to search for a good answer while ensuring that the final answer stays within boundaries that cannot be negotiated?

As AI systems gain greater autonomy, that distinction may become increasingly important. The challenge is not simply to make AI more capable, but to ensure that capability operates within limits that remain under human control.

A search, a payment, a video stream or an AI query may take seconds. Behind it, thousands of machines can be running continuously, drawing power, producing heat and requiring cooling. That is the physical reality of the digital economy. And in India, it is expanding much faster than the size of the facilities themselves might suggest.

India’s installed data centre power capacity has risen from around 375 MW in 2020 to 1.57 GW in August 2026. The government expects it to reach nearly 8 GW by 2030. At the same time, the Central Electricity Authority estimates that data centres could require 17 GW of electricity by 2031–32. The numbers are moving on two different scales: capacity is growing rapidly, while the electricity demand associated with it could grow even faster.

That makes India’s data centre expansion triggers conversations about power, water and infrastructure.

Data Centres: Digital Service and Physical Footprint

The basic operation is familiar. A request from a phone or computer travels through a network to a data centre. Security systems screen it, a load balancer sends it to an available server, the application processes it and a database supplies the required information. The answer then travels back to the user.

What makes the infrastructure demanding is that the machines cannot simply be switched off when demand falls. Servers need continuous electricity. UPS systems provide immediate backup during interruptions and generators provide additional resilience during longer outages. At the same time, cooling systems have to remove the heat generated by the computing equipment.

As computing becomes more intensive, particularly with AI, this underlying infrastructure becomes increasingly important. India’s data centre expansion is therefore not happening in isolation. It is arriving alongside a much broader push into artificial intelligence, semiconductors, cloud computing and domestic IT hardware.

The government itself describes these initiatives as drivers of data-centre growth. The IndiaAI Mission has an outlay of ₹10,371.92 crore, while Semicon 2.0 has been approved with an outlay of ₹1,27,500 crore. Support for the Electronics Components Manufacturing Scheme has also increased from ₹22,000 crore to ₹40,000 crore. More computing, naturally, means more infrastructure to run it.

Electricity Requirement is Where the Scale Becomes Clearer

The government’s 17 GW projection is the figure that deserves closer attention. India’s present data-centre power capacity is 1.57 GW. The projected electricity requirement of 17 GW by 2031–32 is more than ten times that installed capacity figure. These are not directly comparable measures—installed capacity and electricity demand are different metrics—but together they show how rapidly the sector’s energy requirements are expected to expand.
The pressure is not only about producing enough electricity.

Data centres need reliable power around the clock. That makes them a different kind of electricity consumer from facilities whose operations can be shifted to periods of lower demand.

The government’s response is to connect the sector with India’s clean-energy expansion. It points to green open access, the Green Energy Corridor, solar programmes and green hydrogen, while the SHANTI Act is presented as opening a possible role for nuclear power in supplying reliable clean energy to AI and data-centre infrastructure. How much of the coming demand can actually be supplied with clean, reliable electricity?

Adding renewable capacity does not by itself guarantee uninterrupted renewable power at the precise time a server needs it. The answer will depend on transmission, storage, grid management and firm power as much as on new generation capacity.

Water Constraint

The electricity requirement is easier to quantify. The water requirement is harder to see. Cooling technology determines how much water a data centre uses. The government acknowledges this directly, noting the role of direct-to-chip liquid cooling, adiabatic cooling, immersion cooling and closed-loop systems. It also notes that groundwater extraction is subject to regulation.

But the backgrounder does not provide a national estimate of current or projected water consumption by India’s data centres. That omission becomes more significant as the sector expands.

An 8 GW data-centre industry will not have a uniform environmental footprint. Water demand will depend on the technology used, local climate and the source of water. A facility operating in a water-stressed urban region presents a very different resource question from one using recycled water in a water-abundant location. This makes the geography of the data-centre boom important. Where India builds these facilities may matter almost as much as how many it builds.

India Moving Up the Global Data-centre Ladder

The expansion is being driven by more than government policy. India’s huge digital user base, growing cloud adoption and emerging AI demand make it an increasingly attractive location for large data centre operators.

Industry estimates cited in recent reporting put India’s live data-centre capacity at around 1.7 GW, with another 1.3 GW under construction and 3.2 GW in projects that have secured land, power and approvals. Another estimate places India’s colocation capacity at about 2 GW and projects it could reach 10 GW by 2031.

The precise numbers vary depending on how capacity is defined, but the direction is consistent: India is entering the next phase of the global data-centre buildout. That is also reflected in the investment pipeline. The government says nearly $70 billion is already under investment, with another $90 billion in announced projects.

Incentives are Accelerating the Buildout

Policy is helping reduce the barriers to expansion. Data centres received infrastructure status in 2022, giving them greater access to credit. The 2026–27 Budget has also introduced a tax holiday until 2047 for eligible foreign cloud-service providers using India-based data-centre infrastructure.

For India, there are obvious strategic benefits. Domestic capacity can support cloud services, digital payments, e-governance and AI while strengthening the country’s ambitions around data sovereignty and technological self-reliance. But incentives also mean that the public policy question cannot stop at attracting investment.

If data centres receive easier financing and tax advantages while depending on public electricity networks, transmission infrastructure and increasingly scarce natural resources, the economic benefits need to be considered alongside those costs. The release does not provide that calculation.

Efficiency Can Slow the Problem, Not Necessarily Stop It

India does have a framework for improving efficiency. BIS standards cover Power Usage Effectiveness, Carbon Usage Effectiveness, Cooling Efficiency Ratio and Water Usage Effectiveness. BEE’s building codes also contain energy- and water-efficiency provisions relevant to data-centre infrastructure.

These measures matter. More efficient cooling, better server utilisation and lower power overhead can reduce the resources required for each unit of computing. But there is a basic arithmetic problem. If computing demand grows faster than efficiency improves, total resource consumption can still rise.

That is why India’s data-centre story should not be judged solely by whether individual facilities become greener. The larger question is whether the country’s overall expansion is being planned around the limits of its electricity, water and transmission systems.

Next phase: India’s Test Planning

The government describes data centres as foundational infrastructure for a digital economy increasingly shaped by AI and cloud computing. It also acknowledges that their growth requires responsible governance of energy, technology and resources.

India knows how quickly data-centre capacity is growing. It has projections for electricity demand. It is building policies to attract investment and encouraging cleaner energy.

What remains less clear is whether power availability, water availability, grid capacity and environmental limits are being considered together when new facilities are planned.

That is the bigger story behind India’s data-centre boom. The country is not merely building infrastructure for the internet. It is building the physical foundation for an economy increasingly dependent on computation. And the more India moves its economy into the cloud, the more important it becomes to account for what keeps that cloud running.

AI in education is changing the teacher’s role from delivering information to developing critical thinking, ethical judgment, resilience and human connection. As AI becomes a powerful learning tool, teachers may become more important as mentors and guides.

Every few decades, technology forces higher education to step in front of an existential mirror. When search engines made the world’s information queryable in milliseconds, critics prematurely declared the traditional classroom obsolete. Today, as Generative AI (Artificial Intelligence) tools evolve from search interfaces into systems capable of explaining corporate finance models, writing complex code, and breaking down multi-variable calculus step-by-step, we face a far more provocative question: If AI can explain everything, what is actually left for a teacher to teach?

AI in Education Is Changing What Teachers Need to Teach

For decades, higher education was built on an implicit assumption that the teacher’s primary job was the delivery of content. The educator was the traditional authority figure holding access to knowledge and transmitting it to passive recipients. But if education is reduced merely to information delivery, AI will indeed render the traditional teacher redundant. As academic leaders navigating a multidisciplinary university environment, we know that information is not wisdom, and an explanation is not true understanding. As AI becomes a ubiquitous, hyper-efficient personal tutor, the role of the teacher is not shrinking; it is undergoing a profound and necessary elevation.

Generative AI models are exceptional at providing fast answers, but they are fundamentally indifferent to truth, context, and nuance. They synthesise probabilistic patterns rather than human meaning. In a time when students can produce a well-written study of a business case or legal precedent in a matter of seconds, the main responsibility of educators is to teach students how to evaluate answers rather than how to obtain them. In the era of artificial intelligence, critical thinking is now about rigorous investigation rather than knowledge retrieval. Instructors must teach students how to analyse AI outputs, including how to spot algorithmic bias, find logical fallacies, confront minute errors, and comprehend the moral trade-offs and human costs associated with important choices. An algorithm cannot match the human mind’s ability to navigate ambiguity, which is necessary to develop this level of critical judgment.

Higher education is also a very emotional and social process. Vulnerability is necessary for learning; a pupil must first own their ignorance in order to acquire new information. While an AI can adjust to a student’s cognitive tempo, it is unable to recognise the faint spark of inquiry in a quiet seminar room, read the exasperation on a student’s face, or detect a sudden lack of confidence before a significant presentation. Teachers work as mentors, fostering character, ethical responsibility, and resilience. They foster collaborative environments where leadership thrives, set an example of sensitivity, and encourage discussion among peers who hold different opinions. A machine can outline the technical concepts of corporate governance or bioethics, but only a human mentor can instil the moral compass required to lead responsibly in an unpredictable world.

If AI can explain the what and the how, the teacher remains the master of the why.

Instead of replacing teachers, AI presents a unique opportunity by relieving them of repetitious explanations and mundane administrative duties. Teachers might recover their primary role as learning architects by delegating foundational drills and basic synthesis to automated technologies. Project-based challenges, real-world simulations, and interdisciplinary discussions where human collaboration is unavoidable are replacing passive lectures in forward-thinking educational institutions. In order to help students identify their unique skills, intellectual interests, and sense of agency, they concentrate on posing open-ended questions that lack tidy, algorithmic solutions.

As we evaluate the future of our educational institutions, we should not view AI as a competitor to the teacher, but as a clarity-enforcing mirror. If AI can explain the what and the how, the teacher remains the master of the why. As technology democratises access to knowledge at an unprecedented scale, the human touch in education becomes more valuable, not less. The future belongs not to machines that hold all the answers, but to human beings who can think critically, act ethically, collaborate empathetically, and lead with purpose. Guiding young minds to that destination is something only a teacher can do.