Agentic AI reshapes cybersecurity beyond the hype – News

Artificial intelligence is driving a new frontier in cybersecurity, with agentic AI emerging as the next major leap beyond last year’s large language model hype. This evolving technology promises autonomous, goal-driven decision-making — but it also raises new challenges in oversight, trust and responsible deployment.

While early adopters see game-changing potential in streamlining threat detection and response, security leaders are quick to point out that speed without accountability can create fresh vulnerabilities. They stress the need for clear governance frameworks, transparent model behavior and shared responsibility between vendors and customers. In their view, the real opportunity lies not just in building smarter AI agents, but in ensuring they operate within guardrails that protect both systems and reputations, according to Mike Nichols (pictured), vice president of product management at Elasticsearch B.V.

“Everything is agentic now, it was AI last year, now it’s agentic AI this year,” Nichols said. ” I wonder if people know what it means when they’re asking for it. I would hate to be in the customer’s shoes right now, trying to hear the reality.”

Nichols spoke with theCUBE’s Jackie McGuire at the Black Hat USA event, during an exclusive broadcast on theCUBE, News Media’s livestreaming studio. They unpacked the reality behind today’s AI buzzwords, the importance of agentic models over LLM hype and why human analysts remain irreplaceable in modern security operation centers. (* Disclosure below.)

From AI hype to Agentic AI reality

The AI hype train is somewhat stalled at the moment, with numerous prototypes and flashy product demos that show little practical utility. Thus, the rise of agentic AI could serve as a much-needed course correction. Unlike monolithic LLMs trained on noisy internet data, agentic models leverage a suite of AI tools — regression, deterministic and probabilistic models — each selected for its specific task. This allows for more control, accuracy and transparency, according to Nichols.

“There’s definitely some good stuff out there,” he said. “But I think, as always, you have to sift through it all to figure out where there are some beneficial things that are in the environment.”

Despite growing fears that AI might eliminate entry-level analyst roles, the opposite scenario will eventually ring true, according to Nichols. Rather than replace human talent, AI should augment it — automating mundane tasks such as querying logs or accessing siloed systems. This frees up analysts to do what they do best: critical thinking, pattern recognition and decision-making.

“One of the best SOC analysts I ever met was a meteorologist who just understood weather patterns and brought that data science in,” Nichols said. “I think we can get more of those people and break the barrier of who can be in the security team, which would be great … diversify the entire place. But I think AI’s going to allow us to get more humans involved, not remove humans. I don’t think that’s a good concept to follow.”

Here’s the complete video interview, part of News’s and theCUBE’s coverage of the Black Hat USA event:

(* Disclosure: Elasticsearch B.V. sponsored this segment of theCUBE. Neither Elastic nor other sponsors have editorial control over content on theCUBE or News.)

Photo: News