Agentic AI widening cyber risks even as defence tools improve: Trellix MD

Frontier AI models are rapidly improving vulnerability detection but also giving threat actors greater scale, making contextual intelligence and faster action critical, warns Trellix's Ganesh Iyer

A basic action gap persists in cybersecurity for companies despite the availability of intelligence, tools, and services powered by frontier artificial intelligence (AI) models, says Ganesh Iyer, managing director for Trellix (India and SAARC), a global cybersecurity company that was created after the merger of McAfee Enterprise and FireEye.

 

To fill this gap, companies need to work with the right cybersecurity vendors who can help them identify which areas to focus on and which network vectors need to either be replaced or changed, he said.

 

Citing the example of cybersecurity issues that have cropped up due to the advent of frontier AI models such as Anthropic’s Claude Mythos, Iyer said that the past six months have been a “leapfrog of generations” in terms of networks' vulnerability and data security of companies.

 

“These vulnerabilities were not there by design, but it was just something that got missed. But then there was this frontier AI model which was able to figure it out and tell you what problems existed where in just about six hours,” Iyer said, adding that this had also resulted in a multiplier effect for AI-led threat actors.

 

The advent of newer frontier models that can find decades-old vulnerabilities in networks and companies’ data structures has also given rise to threat actors that now offer malware-as-a-service or cybersecurity-threat-as-a-service, he said.

 

“Our advanced research group is, at any given point in time, tracking 500-600 such known actors, some of which are nation-state actors. Together they have run some 5,500 campaigns, which we have been able to track on a regular basis by running them against 4,800 parameters that check for malicious and non-malicious tools and services,” Iyer said.

 

The use of AI in cybersecurity is therefore moving toward a situation where both companies and bad-faith actors will leverage AI to the hilt, with the difference coming in the better contextualisation of the intelligence available to either party, he said.

 

“The fastest finger right now is on the AI button only, especially on the agentic part of it because there are increasing amounts of examples where an agent went rogue, or it was able to assume a different identity. We are still scratching the surface when it comes to the cybersecurity impact of AI agents,” Iyer said.

 

Overall, the cybersecurity market, even in developing countries such as India, will grow at least 30-40 per cent over the next six to eight months as companies take more seriously protecting themselves against agentic AI-led threats, he said.