Advance Cybersecurity and Resilience Maturity for AI with Dell Technologies

Steve Kenniston of Dell Technologies presents how organizations can advance cybersecurity and resilience maturity for AI. The session covers governance and funding for AI security, agentic AI as a growing attack surface, Dell’s three practice areas of reducing the attack surface, detecting and responding to threats, and recovering operations, along with supply chain security, post-quantum readiness, and using AI for defense. Kenniston, presenting at Security Field Day 16, framed the session around the surge of AI security questions Dell now hears from customers.

Kenniston said attackers inside AI environments now break out in about 29 minutes, and 41 percent of organizations surveyed say they need better security and resilience before they can scale AI. While 82 percent trust their existing policies, only one in five can see what their agents are actually doing, and roughly 80 percent of attacks now use legitimate identities. He urged teams to build security into AI project budgets rather than handing it to the security team at the last minute. Dell applies its three practice areas to each product, illustrated by PowerProtect, which pairs MFA, role-based access, and dual authorization with CrowdStrike and Index Engines detection, plus vaulting and clean room recovery. Silicon root of trust, secure component verification, and Fortanix confidential computing extend that protection from the supply chain into memory.

Kenniston stressed that security usually fails in the seams between products, positioning Dell’s partners as the grout that closes those gaps. For agentic AI, he recommended low initial autonomy, approval gates, sandboxes, and human oversight to contain destructive actions, runaway token loops, and unintended outreach. Dell is shipping quantum-resistant commercial PCs but won’t call a system quantum resistant until every component inside it, including partner NICs and HBAs, qualifies. On defense, Dell is using frontier models to find its own flaws first and has signed an agreement with more than 100 partners to widen defender access to those models. His closing advice was practical: measure actual recovery time against leadership’s expectation, give every agent an identity, and treat the gap between the two as the roadmap.