AI Agents Create New Enterprise Security Risks Beyond Traditional Defenses

AI workloads transform the enterprise attack surface

As enterprises deploy artificial intelligence at scale, they're creating what analysts call "AI factories"—systems that continuously generate, transform, and consume data across distributed environments. These factories produce intelligence in the form of tokens, but they also introduce security vulnerabilities that traditional cybersecurity frameworks weren't designed to address.

Unlike conventional software that behaves predictably, AI models are adaptive and probabilistic. They evolve over time, and increasingly, AI agents act autonomously—interacting with enterprise systems, executing workflows, and making decisions with minimal or no human oversight. This shift fundamentally changes what needs to be protected and how.

Why it matters

The rise of autonomous AI agents represents a security inflection point for enterprises. When agents can take actions independently at machine speed, the consequences of compromise extend beyond data breaches to operational disruption. Organizations that fail to build identity governance and hardware-level security into their AI infrastructure risk losing control over the intelligence they produce.

Autonomous agents demand new identity controls

AI agents present a distinct security challenge because they can take actions, not just provide answers. According to Dell's Mukund Khatri, fellow and vice president of systems architecture, while large language models might give wrong answers, agents can take wrong actions—a more serious problem when systems operate autonomously in real time.

The scale of concern is significant. Darktrace's 2026 State of AI Cybersecurity report, which surveyed 1,540 cybersecurity leaders across 14 countries, found that 92% were concerned about security implications of AI agents in their workforce. Nearly half—46%—said they weren't adequately prepared to defend against AI-powered threats. The report also revealed that 87% believe AI is significantly increasing malware sophistication and success rates.

Identity has become a critical control point. Organizations need to know which agents are operating, what they can access, what actions they can take, and when permissions should expire. Krista Case, principal analyst at theCUBE Research, notes that extending governance concepts like lifecycle management and least-privilege access from human identities to agents doesn't require starting from scratch—the frameworks exist, but they must adapt to a rapidly growing population of non-human identities.

Steve Kenniston, senior cybersecurity evangelist at Dell, reports that 85% to 90% of AI implementation projects get stopped because security teams weren't involved early enough. This underscores the need for security involvement before agentic systems begin interacting with sensitive data and business processes.

Hardware forms the security foundation

Dell and Intel are addressing AI security by building protection into infrastructure rather than layering it on afterward. Dell integrates security from supply chain through chips to delivered devices, embedding roots of trust in components and using cryptographic image signing and SHA-384 hash verification to verify AI model container integrity before deployment.

Intel's Mike Ferron-Jones, go-to-market lead for platform security and integrity, emphasizes that the CPU serves as the fundamental hardware root of trust. Intel organizes its data center security capabilities into four areas: platform protection, confidential computing through SGX and TDX, software behavior enforcement, and encryption acceleration.

Confidential AI environments place workloads inside trusted execution environments with hardware-enforced isolation and cryptographic attestation. Intel has developed reference architecture with Nvidia that combines CPU trusted execution environments with GPU confidential-computing capabilities, extending protection to GPU-accelerated workloads.

Post-quantum cryptography enters production roadmaps

Both companies are preparing for quantum computing threats. Intel expects all cryptographic operations inside its platforms to use quantum-safe technology by 2029, with full post-quantum cryptography compliance across new platforms by 2030. Dell is similarly moving toward quantum-resistant protections across its portfolio.

The "harvest now, decrypt later" scenario—where attackers collect encrypted data today to decrypt once quantum computers become capable—is driving urgency around post-quantum readiness.

As Dave Vellante, chief analyst at theCUBE Research, notes, security must become the control plane that governs how intelligence is produced. "If you cannot secure the AI factory, you do not control the outcome," he said.

These details were first reported by SiliconANGLE in its coverage of the "Securing the AI Factory With Dell Technologies and Intel" event.