In a first, Karnataka cops to deploy agentic AI to combat fake news

The Karnataka police are set to deploy a cutting-edge artificial intelligence system to combat fake news, cybercrimes and online misinformation. 

A few years ago, the state launched the Information Disorder Tackling Unit (IDTU) under the Department of Electronics, IT and BT to address the growing threat of fake news. That unit has now been integrated into the newly formed Cyber Command Unit (CCU) under the Home Department. 

At the heart of the CCU will be the planned deployment of agentic AI systems — artificial intelligence designed to independently scan the internet for harmful content and flag it for review. If the content is deemed to have broader public relevance, particularly in the case of misinformation, it will be published on an interactive government platform called satya.gov.in. 

DH has reviewed all documents related to this initiative. 

What Is agentic AI?

Unlike traditional AI models that rely heavily on human supervision, agentic AI systems can perform complex tasks autonomously and make real-time decisions. They function like digital assistants working for police officers, according to a senior police official involved in the project. 

"These agents are deployed across public parts of the internet to identify and flag harmful content. They do not access private messages or personal conversations," the officer added. 

Once flagged, the content is reviewed by human operators. As the system matures and its accuracy improves, these AI agents will also be able to make certain decisions — such as removing harmful content — on their own. 

In addition, when misinformation that could impact the public is detected, the AI will reframe it in a simple, easy-to-understand format and post it on satya.gov.in. Citizens will also be able to contribute by sharing verified information or flagging misleading content. 

To make agentic AI a reality, the system needs to be trained on massive volumes of data. To facilitate this, police have begun working with Large Language Models (LLMs) and Small Language Models (SLMs) — machine learning systems capable of handling specific tasks with minimal input. 

"These models are currently being used to complete targeted tasks. Over time, as we gather more data, we'll use it to train our agentic systems," the officer said. 

All development work is being carried out in-house at the IDTU. Till recently, the unit had been relying on traditional AI to monitor misinformation online and had achieved strong results with high accuracy. However, those systems were limited in scope and required substantial human oversight. 

"With agentic AI, we aim to reduce that dependency. These systems will eventually handle a larger share of the workload autonomously," the officer explained. 

This sophisticated technology is being developed internally using existing resources. "We have technically skilled officers who are capable of managing this initiative. If required, we'll seek expert consultation but the foundation is being built by our own team," the officer said. 

Challenges remain, however. 

The officer cautioned that even advanced AI systems can occasionally generate false or misleading conclusions — a phenomenon known as "hallucination". Because of this, human oversight will continue even after deployment. 

"Developing and fine-tuning such high-tech systems requires continuous observation, research and testing," the officer added. 

Despite the ambition behind the initiative, Karnataka has allocated only Rs 5 crore for tackling cybercrimes this year. Whether this will be sufficient to support such a technologically intensive project remains to be seen.