Speakers presenting on AI cybersecurity with a digital globe and network visuals behind

Cisco’s Jeetu Patel on overcoming the ‘AI trust deficit’

IT Management Comments Off on Cisco’s Jeetu Patel on overcoming the ‘AI trust deficit’

The management issue is not whether AI can help security teams, but who is accountable when it acts. Agentic tools only become useful when leaders can define the boundary between assistance and delegation. That means clear decision rights for what an AI agent may observe, recommend, block or execute, plus explicit escalation paths when confidence is low or behaviour is unusual.

Observability is therefore a governance requirement, not just a technical feature. If leaders cannot trace model behaviour, runtime context and downstream actions, they cannot prove control, tune risk appetite or satisfy audit and legal scrutiny. The practical question for CIOs and CISOs is which signals must be monitored centrally, which can remain in domain tools, and how exceptions will be reviewed without creating a manual bottleneck.

The operating model also changes. Moving toward machine-speed detection and response can reduce dwell time, but it may also compress the time humans have to validate alerts and approve containment. That creates a trade-off between automation and oversight: faster response versus a higher need for guardrails, testing and role redesign for SecOps, architecture and risk teams. Leaders should ask where human approval remains mandatory and where pre-authorised playbooks are acceptable.

Finally, platform consolidation is a portfolio decision, not merely a procurement one. Integrated controls may improve visibility and reduce integration burden, but they can also deepen vendor dependence and delay best-of-breed choices. The next questions for governance boards are whether the current toolset can support policy consistency across AI agents, how third-party integrations will be risk-assessed, and which metrics will prove that AI is reducing exposure rather than simply increasing activity.

LAS VEGAS โ€“ Anthropic Mythos put a spotlight on how agentic AI is changing the game for cybersecurity, causing enterprises toย rethink how their security teams are structuredย and how they can prevent AI-based threats. “AI changes the speed of defense. The bad corollary to that is it’s empowering our adversaries at a pace that we’ve never seen in our career,” said Chuck Robbins, CEO of Cisco, in a keynote Tuesday at Cisco Live. While AI-based threats are on the rise, there’s also an opportunity to use AI to thwart cybersecurity threats. But in order to do so, enterprises will also need to overcome the “AI trust deficit,” according to Jeetu Patel, president and chief product officer for Cisco. “One of the biggest challenges people have with [AI] agents right now is they don’t trust it, and if you don’t trust it, you’re not going to delegate work to them to use them,” said Jeetu during his keynote Tuesday. Solving the “AI trust deficit” will require observability across the entire AI stack, which includes achieving visibility and insights into application runtime and model performance, Patel explained. That requirement is easier stated than accomplished. “AI agents are kind of like teenagers. They’re supremely intelligent, but they have no fear of consequence, and sometimes they do stupid stuff. So, you need to make sure that you can protect the world from them,” Patel said. Cisco is working with other networking and technology companies to examine how enterprise customers can improve their cybersecurity strategies in the AI era. As part ofย Anthropic’s Project Glasswing, Cisco was among the first 50 open source, networking, and cybersecurity companies tasked with testing the Anthropic Mythos AI model, which can both identify and exploit zero-day vulnerabilities in open-source codebases. By participating in Project Glasswing, “Our goal was to understand how we could safeguard our customers, how we could come together as a broader ecosystem to work on AI-powered defense and to stop AI-powered threats,” Robbins said. He added that AI-based threats have the potential to have far-reaching implications for businesses, governments, and society overall. Anthropic said in statement Tuesday it is extending its Project Glasswing partnership to approximately 150 new organizations.

Three-pronged approach to agentic security

Patel discussed how Cisco is working with enterprises in securing their agentic workforce via three main goals: by protecting AI agents from external threats, preventing AI agents from creating cybersecurity threats, and detecting and responding to threats “at machine speed.” Cisco’s AI threat strategy includes the launch ofย Cisco Cloud Controlย this week, available on a limited basis in the US. It’s currently being used by about 60 organizations including semiconductor company AMD. Cisco Cloud Control expands on last year’s launch ofย Security Cloud Control. The new product is a cloud-based network infrastructure management platform that embeds security services with a number of other IT and networking infrastructure applications โ€“ both Cisco applications and 50 third-party vendors’ applications โ€“ on a single platform. Enterprise customers don’t want to be system integrators anymore, but are looking to vendors to provide these types of integrated platforms, Robbins said during a press and analyst conference after the keynotes. Patel said he believes that โ€“ with AI โ€“ the ability to analyze every security signal/alert is within reach, predicting the emergence of agentic SOCs that can quickly identify and prevent network anomalies. But for SecOps teams to operate at “machine speed and scale,” they will need to invest in network visibility, threat validation, and security guardrails for AI agents, he said.  

Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.