'Fraudsters making your AI go rogue': KPMG India's S Sathish on types of cyber frauds due to AI

Artificial intelligence (AI), just like the internet, is becoming increasingly prone to cyber fraud. S Sathish, Partner & National Leader - Industrial Manufacturing, KPMG in India, said at the Business Today AI Summit 2025 that fraudsters are leveraging AI to create more force to penetrate and create cyber fraud.

"If you really look at the two kinds of fraud that are actually happening. It's one of course, they are leveraging AI for creating more force to penetrate and create a cyber fraud. So it necessarily means organisations also need to respond with that agility to have a very strong digital resilience to go about responding," he said. 

Sathish also explained that there is another type of cyber fraud wherein fraudsters make the company's AI go rogue.

"The second type of cyber fraud which are actually happening is that they are getting into the code of your AI itself and making your AI go rogue. To ensure that, because many of the decisions today are actually taken based on AI." 

He mentioned that if one influences the logic or changes certain things, the decisions taken using artificial intelligence actually end up causing loss. Sathish said that while there is a lot of confidence among companies on AI, it is equally important to develop a very strong digital defence against cyber frauds.

He added that organisations need to have strong digital resilience to respond to such incidents, while mentioning that investment needs to be made into AI security.

Backing the government's initiatives on AI adoption, Sathish said that it should handhold and provide financial support to the MSMEs to tackle challenges posed by AI. He added that the government should coach them on a cluster basis about the risks associated with artificial intelligence.

His comments assume importance as OpenAI is developing an AI system to serve as an intern-level research assistant by September 2026, aiming for a "legitimate AI researcher" by 2028. CEO Sam Altman shared these plans during a livestream, outlining the ambition to move AI beyond generating text and images, towards supporting real research work.

To reach this milestone, OpenAI is advancing algorithms for improved learning and reasoning, and increasing "test-time compute" to provide more computational resources for problem-solving.