AI's dual role in cybersecurity: Threat and defence, says Sridhar Vembu

Zoho's Sridhar Vembu warns that falling AI costs will empower cyber attackers but also presents an opportunity for defenders. He urged India to leverage AI proactively to find vulnerabilities and build 'provably secure systems' from the ground up.

Falling costs of super intelligence (AI) could make sophisticated cyber capabilities more accessible to attackers, but India and other countries can use the same technology to identify vulnerabilities faster and build more secure digital systems, Zoho Corporation Founder and Chief Scientist Sridhar Vembu said.

Speaking on the sidelines of the Indian Army-CyberPeace Terrier Cyber Quest 3.0 Conclave, Vembu said declining AI token costs would benefit both sides of the cybersecurity equation, making it essential for defenders to deploy AI proactively. “The people who attack have cheap tokens to spend on it. People who defend also, we have to find and close vulnerabilities before the bad people find them,” Vembu said.

A Shift from 'Whack-a-Mole' to 'Provably Secure' Systems

He said the challenge goes beyond simply detecting and patching vulnerabilities, pointing to an opportunity to develop technologies in which vulnerabilities can be mathematically or technically demonstrated not to exist. “We also have an opportunity to adopt truly secure technologies where no vulnerabilities can be, we can prove that no vulnerabilities exist that can be exploited,” he said.

Vembu described the traditional cybersecurity model as a continuous cycle of discovering vulnerabilities and fixing them, likening it to a “whack-a-mole game”. Instead, he said, AI could help build “provably secure systems” that reduce the scope for vulnerabilities to be exploited in the first place.

India's Opportunity in Building Secure AI

He said India has an opportunity to build such systems by leveraging its strengths in software development and verification.

Leveraging Strengths in Code and Verification

According to Vembu, large language models (LLMs) generate code, which then passes through verification before being deployed, with runtime feedback subsequently helping regenerate code. “We are very strong in the code and verification part,” he said, adding that India could build tools around this architecture and train cheaper models as AI becomes more commoditised.

Vembu said this could eventually make secure AI systems more affordable for India’s national requirements, while strengthening the country’s ability to protect critical digital infrastructure from increasingly AI-enabled cyber threats. (ANI)

(Except for the headline, this story has not been edited by Asianetnews Editorial staff and is published from a syndicated feed.)