LAS VEGAS – Anthropic Mythos put a highlight on how agentic AI is altering the sport for cybersecurity, inflicting enterprises to rethink how their safety groups are structured and the way they’ll stop AI-based threats.
“AI adjustments the velocity of protection. The unhealthy corollary to that’s it is empowering our adversaries at a tempo that we have by no means seen in our profession,” stated Chuck Robbins, CEO of Cisco, in a keynote Tuesday at Cisco Dwell.
(Supply: Kelsey Ziser/InformationWeek) Cisco CEO Chuck Robbins introduced a keynote at Cisco Dwell 2026 in Las Vegas.
Whereas AI-based threats are on the rise, there’s additionally a possibility to use AI to thwart cybersecurity threats. However so as to take action, enterprises may also want to beat the “AI belief deficit,” in response to Jeetu Patel, president and chief product officer for Cisco. “One of many largest challenges individuals have with [AI] brokers proper now’s they do not belief it, and in the event you do not belief it, you are not going to delegate work to them to make use of them,” stated Jeetu throughout his keynote Tuesday.
Fixing the “AI belief deficit” would require observability throughout the complete AI stack, which incorporates reaching visibility and insights into utility runtime and mannequin efficiency, Patel defined. That requirement is simpler acknowledged than achieved.
“AI brokers are sort of like youngsters. They’re supremely clever, however they haven’t any worry of consequence, and typically they do silly stuff. So, it’s worthwhile to just remember to can shield the world from them,” Patel stated.
Cisco is working with different networking and expertise corporations to look at how enterprise clients can enhance their cybersecurity methods within the AI period. As a part of Anthropic’s Undertaking Glasswing, Cisco was among the many first 50 open supply, networking, and cybersecurity corporations tasked with testing the Anthropic Mythos AI mannequin, which might each determine and exploit zero-day vulnerabilities in open-source codebases.
By collaborating in Undertaking Glasswing, “Our objective was to know how we may safeguard our clients, how we may come collectively as a broader ecosystem to work on AI-powered protection and to cease AI-powered threats,” Robbins stated. He added that AI-based threats have the potential to have far-reaching implications for companies, governments, and society general.
Anthropic stated in assertion Tuesday it’s extending its Undertaking Glasswing partnership to roughly 150 new organizations.
Three-pronged method to agentic safety
Patel mentioned how Cisco is working with enterprises in securing their agentic workforce by way of three most important targets: by defending AI brokers from exterior threats, stopping AI brokers from creating cybersecurity threats, and detecting and responding to threats “at machine velocity.”
Cisco’s AI risk technique contains the launch of Cisco Cloud Management this week, obtainable on a restricted foundation within the US. It is presently being utilized by about 60 organizations together with semiconductor firm AMD.
Cisco Cloud Management expands on final yr’s launch of Safety Cloud Management. The brand new product is a cloud-based community infrastructure administration platform that embeds safety providers with numerous different IT and networking infrastructure purposes – each Cisco purposes and 50 third-party distributors’ purposes – on a single platform.
Enterprise clients do not wish to be system integrators anymore, however want to distributors to supply these kind of built-in platforms, Robbins stated throughout a press and analyst convention after the keynotes.
Patel stated he believes that – with AI – the flexibility to investigate each safety sign/alert is inside attain, predicting the emergence of agentic SOCs that may shortly determine and stop community anomalies. However for SecOps groups to function at “machine velocity and scale,” they might want to put money into community visibility, risk validation, and safety guardrails for AI brokers, he stated.
