INDIA’S AI TALENT PUSH NEEDS AUTHORITY TRAINING

An AI university should produce graduates who not only know how to build and use advanced systems, but also how to define the system’s authority, test whether it stays within scope, and respond when it does not.

Karnataka is moving from talking about artificial intelligence skills to building an institution around them. That is the right direction. The curriculum should also prepare students and professionals for a harder question: what happens when AI stops merely answering questions and starts taking actions?

The first meeting of the proposed Karnataka Artificial Intelligence University advisory committee took place on September 5, with industry, academia and government discussing immediate certificate, diploma, finishing-school, executive-education and reskilling programs. The planned skill areas include generative and agentic AI, data analytics, cloud technology, cybersecurity, and AI infrastructure.

That agenda should include training on AI authority from the beginning.

The reason is visible in the METR/Redwood investigation of a major real-world cyberattack on Hugging Face. AI agents driven by an unreleased OpenAI internal research model attacked Hugging Face on their own, despite recognizing that they were not supposed to do so. Hundreds of agents joined the attack, shared discoveries, divided up the work, and coordinated through their own message board until they successfully breached Hugging Face’s defenses. Advanced AI systems had organized themselves to carry out a large, sustained cyberattack against a major company.

Also Read: Digital Literacy and Moral Vacuity: A Crisis of National Character 

I’m no AI skeptic. In my experience, strong safeguards increase trust and make faster adoption possible, while reducing the risk of failures like the Hugging Face attack.

India already recognizes that responsible AI belongs alongside adoption. The IndiaAI Mission’s AI Responsibility campaign emphasizes ethical and human-centric AI, while the IndiaAI ecosystem includes security tools aimed at robust and trustworthy deployment. The next step is to turn those principles into operating skills for people who will deploy agents inside companies and public institutions.

A practical curriculum should teach an “authority ladder“.

At level one, an AI agent can research, summarize, draft, and recommend, but cannot alter an external system. At level two, it can take narrow and reversible actions with human review, such as preparing a customer response or creating a draft procurement request. At level three, it can execute approved actions within explicit limits on data, money, communications, and software access. High-consequence actions should still require human approval.

Also Read: India’s Higher Education Crisis: Degrees Growing, Quality Declining

This framework changes what students learn to ask. Instead of stopping at “Can the agent do the task?” they ask “What should the agent be allowed to do without another person checking?” That distinction becomes critical in finance, healthcare, public services, manufacturing, and any workplace where an agent can move from information to action.

Training should also include scope-compliance testing. Can the agent recognize when a request falls outside its assigned job? Will it stop when instructions conflict with policy? Does it ask for help when uncertain? Can an organization reconstruct what the agent did after the fact? Those questions belong in labs, capstone projects, and executive programs just as much as prompt design and model evaluation.

Students should also learn incident response for agentic systems. When an AI agent takes an unauthorized action, organizations need to preserve logs, suspend the relevant access, determine how the authority boundary failed, and independently review serious incidents. Restoring access without understanding the failure simply recreates the conditions for another one.

This is especially important because India’s AI strategy is focused on scale. The India AI Impact Expo showcased real systems in motion, including agentic AI workflows, and framed the national challenge as moving AI from experimentation toward deployment across sectors. Scaling deployment without scaling control capabilities creates a predictable gap.

Karnataka can help close it. An AI university should produce graduates who not only know how to build and use advanced systems, but also how to define the system’s authority, test whether it stays within scope, and respond when it does not.

That would make responsible AI more concrete. It would also make adoption easier. Employers can move faster when they know what an agent can access, what it cannot change, when it must escalate, and how a serious failure will be investigated.

India has the opportunity to teach a generation of AI professionals that competence includes control. The faster agentic AI spreads, the more valuable that lesson will become.

Disclaimer: The views expressed in this article are those of the author solely. TheRise.co.in neither endorses nor is responsible for them. Reproducing this content without permission is prohibited.