Discover the event trusted by enterprise leaders for almost two decades. VB Transform is the gathering place for those shaping real enterprise AI strategy. Find out more
Stolen credentials continue to be the root cause of 80% of enterprise breaches. The consensus among major security vendors is clear: Identity now serves as the control plane for AI security. The sheer scale of operations necessitates this shift. With enterprises overseeing 100,000 employees, managing over a million identities when AI agents come into play becomes a reality.
Conventional identity access management (IAM) architectures are ill-equipped to handle the surge in agentic AI proliferation. Originally designed for human users in the thousands, these systems struggle to secure millions of autonomous agents operating at machine speed with human-level permissions. The industry’s response marks a significant security transformation since the inception of cloud computing.
Proximity-based authentication replaces hardware tokens
Leading vendors have adopted Bluetooth Low Energy (BLE) technology between devices and laptops to verify physical proximity. This, coupled with cryptographic identities and biometrics, establishes a four-factor authentication system without the need for hardware tokens.
Cisco’s Duo showcases this innovation at scale. Their proximity verification offers phishing-resistant authentication through BLE-based proximity combined with biometric verification. Introduced at Cisco Live 2025, this capability marks a fundamental shift in authentication architecture.
Microsoft’s Entra ID successfully manages 10,000 AI agents in single pilot programs while handling 8 billion authentications daily. “Traditional directory services were not designed to cater to autonomous systems operating at this velocity,” notes Alex Simons, CVP of identity at Microsoft.
Ping Identity’s DaVinci orchestration platform goes a step further. Processing over a billion authentication events daily, with AI agents constituting 60% of the traffic, each verification is completed in under 200 milliseconds while maintaining cryptographic integrity.
Behavioral analytics catch compromised agents in real time
CrowdStrike treats AI agents as any other identity threat. Their Falcon platform establishes behavioral baselines for each agent within 24 hours, triggering automated containment upon deviations within seconds.
“When an AI agent deviates from its established pattern by accessing systems outside the norm, we treat it just like a compromised employee credential,” explains Adam Meyers, head of counter adversary operations at CrowdStrike. The platform monitors 15 billion AI-related events daily across customer environments.
The swiftness of response is crucial. CrowdStrike’s 2025 Global Threat Report reveals that adversaries achieve initial access in under 10 minutes, moving laterally across 15 systems within the first hour. AI agents operating with compromised identities exponentially amplify the damage.
Identity resilience prevents catastrophic failures
On average, enterprises have 89 different identity stores across cloud and on-premises systems, according to Gartner. This fragmentation creates vulnerabilities that adversaries exploit regularly. The solution lies in applying networking principles to identity infrastructure.
Okta’s Advanced Server Access incorporates redundancy, load balancing, and automated failover mechanisms across identity providers. In the event of primary authentication failure, secondary systems activate within 50 milliseconds, a critical feature when AI agents execute thousands of operations per second.
“Identity is security,” asserts Todd McKinnon, CEO of Okta, at Oktane 2024. “When AI is operationalized, granting agents access to real systems and data, a compromised agent identity can have cascading effects across millions of automated actions.”
Zero trust scales for agent proliferation
Palo Alto Networks’ Cortex XSIAM completely shifts away from perimeter defense, operating on the assumption of continuous compromise. Every AI agent undergoes verification before each action, not just during initial authentication.
Mike Riemer, Field CISO at Ivanti, underscores the zero trust approach in a recent interview with VentureBeat: “It operates on the principle of ‘never trust, always verify.’ By embracing a zero trust architecture, organizations can ensure that only authenticated users and devices access sensitive data and applications.”
Cisco’s Universal ZTNA extends this model to AI agents, broadening zero trust beyond humans and IoT devices to encompass autonomous AI systems, offering automated discovery and delegated authorization at scale.
Automated playbooks respond instantaneously to identity anomalies. When malware triggers authentication irregularities, XSIAM revokes access and initiates forensic analysis without human intervention, establishing zero-latency response as the standard operating procedure.
Zscaler CEO Jay Chaudhry identifies the fundamental vulnerability at Zenith Live 2025: “Network protocols were designed to facilitate communication between trusted devices. AI exploits this legacy architecture on a massive scale, with adversaries launching phishing campaigns that compromise agent identities faster than human response times allow.”
Universal ZTNA frameworks enable million-agent deployments
The architectural imperatives are evident. Universal zero trust network access (ZTNA) frameworks in the industry offer four essential capabilities crucial for AI environments.
Cisco’s implementation showcases the necessary scale. Their Universal ZTNA platform conducts automated discovery scans every 60 seconds, cataloging new AI deployments and permission sets to eliminate the blind spots that attackers exploit. Cisco’s delegated authorization engine enforces least-privilege boundaries through policy engines processing 100,000 decisions every second.
Robust audit trails document every agent action for forensic scrutiny. Security teams utilizing platforms like Cisco’s can reconstruct incidents across millions of interactions, while native support for standards like the Model Context Protocol ensures interoperability as the ecosystem evolves.
Ivanti’s approach complements these capabilities with AI-driven analytics. Daren Goeson, SVP of product management at Ivanti, emphasizes, “AI-powered endpoint security tools can analyze vast amounts of data to detect anomalies and predict potential threats faster and more accurately than any human analyst, offering clear visibility across devices, users, and networks to proactively identify security gaps.”
Cisco’s AI security architecture sets industry direction
Cisco’s AI Secure Factory positions them as the first non-Nvidia silicon provider in Nvidia’s reference architecture. By integrating post-quantum encryption with new devices, Cisco is constructing infrastructure to safeguard against future threats. The message for enterprises is clear: Securing AI is not a choice; it’s a fundamental architectural requirement.
At Cisco Live 2025, the company unveiled a comprehensive identity and AI security strategy that addresses every layer of the stack:
| Announcement | Core problem solved / strategic value | Technical details | Availability |
| Hybrid mesh firewall (incl. HyperShield) | Distributed, fabric-native security; shifts security from the perimeter into the network fabric | eBPF-based enforcement; hardware acceleration | New firewalls: Oct 2025 |
| Live protect | Reduces “45-day patch vs. 3-day exploit” gap with rapid, kernel-level vulnerability shielding | Real-time patching without reboots | Nexus OS: Sept 2025 |
| Splunk: Free firewall log ingestion | Cuts SIEM costs by up to 80%; promotes Cisco firewall adoption | Unlimited log ingestion from Cisco firewalls | Aug 2025 |
| Splunk: Observability for AI | Offers critical visibility into AI stack performance | Monitors GPU utilization and model performance | Sept 2025 |
| Duo IAM | Evolves from MFA to a comprehensive security-first IAM platform | User Directory, SSO, Identity Routing Engine | Available Now |
| Duo: Proximity verification | Provides phishing-resistant authentication without hardware tokens | BLE-based proximity, biometric verification | Part of the new Duo IAM |
| Duo: Identity resilience | Mitigates critical IDP outage risks | Redundancy, load balancing, and automated failover | In development |
| Cisco universal ZTNA | Extends zero trust to humans, IoT/OT devices, and AI agents | Automated discovery, delegated authorization | Ongoing evolution |
| Open-sourced security AI model | Democratizes AI defense; matches 8B parameters for 70B model performance | Operates on CPU; 5B security tokens training | Available (Hugging Face) |
| AI defense and Nvidia partnership | Secures AI development pipeline | Nvidia NIM microservices optimization | Available now |
| Post-quantum security | Future-proof against quantum attacks | MACsec and IPsec encryption | New devices (June 2025) |
| Identity intelligence | Enables continuous behavioral monitoring | AI-powered anomaly detection | Part of Security Cloud |
| Secure access | Converges VPN and ZTNA capabilities | Cloud-delivered secure access service edge | Available now |
Cross-vendor collaboration accelerates
The Cloud Security Alliance Zero Trust Advancement Center now boasts the participation of every major security vendor. This unprecedented collaboration facilitates unified security policies across platforms.
“Security vendors must unite against common threats,” emphasizes George Kurtz, CEO of CrowdStrike, during a recent platform strategy discussion. “A data-centric approach prevails given the rapid evolution of adversaries and threats.”
Cisco President and CPO Jeetu Patel echoes this sentiment in an interview with VentureBeat: “Security serves as a prerequisite for AI adoption. Without trust in the system, users will not engage with it.”
The key organizational hurdle remains. Robert Grazioli, CIO at Ivanti, identifies CISO and CIO alignment as the critical barrier in 2025. This collaboration proves essential in effectively safeguarding modern businesses. Executives must consolidate resources — budgets, personnel, data, and technology — to enhance the organization’s security posture.
The identity reckoning
When Cisco, Okta, Zscaler, Palo Alto Networks, and CrowdStrike independently arrive at the same conclusions regarding identity architecture, it signifies affirmation, not coincidence.
Identity infrastructure dictates security outcomes. Organizations face a choice: Architect identity as the control plane or resign to breaches as an inevitable reality. The gap between AI deployment speed and identity security maturity diminishes by the day.
Three critical actions cannot be delayed. Audit every AI agent’s identity and permissions within 30 days. Implement continuous verification for all non-human identities immediately. Establish round-the-clock identity security operations to prevent adversaries from exploiting vulnerabilities.
The unified stance among vendors communicates a resounding message. Identity has emerged as the control plane for AI security. Enterprises failing to adapt will find themselves managing breaches instead of driving innovation in 2025.
