Stolen credentials are answerable for 80% of enterprise breaches. Each main safety vendor has converged on the identical conclusion: Id is now the management airplane for AI safety. Scale alone calls for this shift. Enterprises managing 100,000 staff will deal with multiple million identities when AI brokers enter manufacturing.
Conventional identification entry administration (IAM) architectures can’t scale to safe the proliferation of agentic AI. They had been constructed for hundreds of human customers, not hundreds of thousands of autonomous brokers working at machine velocity with human-level permissions. The business response represents essentially the most vital safety transformation for the reason that adoption of cloud computing.
Proximity-based authentication replaces {hardware} tokens
Main distributors now use Bluetooth Low Power (BLE) between units and laptops to show bodily proximity. Mixed with cryptographic identities and biometrics, this creates four-factor authentication with out the necessity for {hardware} tokens.
Cisco’s Duo demonstrates this innovation at scale. Their proximity verification delivers phishing-resistant authentication utilizing BLE-based proximity at the side of biometric verification. This functionality, unveiled at Cisco Dwell 2025, represents a basic shift in authentication structure.
Microsoft’s Entra ID handles 10,000 AI brokers in single pilot packages whereas processing 8 billion authentications every day. “Conventional listing companies weren’t architected for autonomous techniques working at this velocity,” states Alex Simons, CVP of identification at Microsoft.
Ping Id’s DaVinci orchestration platform pushes additional. The system processes greater than 1 billion authentication occasions every day, with AI brokers accounting for 60% of the visitors. Every verification completes in beneath 200 milliseconds whereas sustaining cryptographic proof.
Behavioral analytics catch compromised brokers in actual time
CrowdStrike treats AI brokers like some other identification menace. Their Falcon platform establishes behavioral baselines for every agent inside 24 hours. Deviations set off automated containment inside seconds.
“When an AI agent immediately accesses techniques outdoors its established sample, we deal with it identically to a compromised worker credential,” Adam Meyers, head of counter adversary operations at CrowdStrike, advised VentureBeat. The platform tracks 15 billion AI-related occasions every day throughout buyer environments.
That velocity issues. CrowdStrike’s 2025 International Menace Report paperwork that adversaries are attaining preliminary entry in lower than 10 minutes. They transfer laterally throughout 15 techniques throughout the first hour. AI brokers working with compromised identities amplify this harm exponentially.
Id resilience prevents catastrophic failures
Enterprises common 89 completely different identification shops throughout cloud and on-premises techniques, based on Gartner. This fragmentation creates blind spots that adversaries exploit every day. The repair applies networking ideas to identification infrastructure.
Okta’s Superior Server Entry implements redundancy, load balancing and automatic failover throughout identification suppliers. When main authentication fails, secondary techniques activate inside 50 milliseconds. This turns into necessary when AI brokers execute hundreds of operations per second.
“Id is safety,” Todd McKinnon, CEO of Okta, stated at Oktane 2024. “Whenever you transfer AI into manufacturing, you give brokers entry to actual techniques, actual information and your buyer information. One compromised agent identification cascades throughout hundreds of thousands of automated actions.”
Zero belief scales for agent proliferation
Palo Alto Networks’ Cortex XSIAM fully abandons perimeter protection. The platform operates on the idea of steady compromise. Each AI agent undergoes verification earlier than every motion, not simply at preliminary authentication.
Mike Riemer, Area CISO at Ivanti, strengthened the zero belief method in a current interview with VenturBeat: “It operates on the precept of ‘by no means belief, at all times confirm.’ By adopting a zero belief structure, organizations can be sure that solely authenticated customers and units acquire entry to delicate information and purposes.”
Cisco’s Common ZTNA extends this mannequin to AI brokers. The platform expands zero belief past people and IoT units to embody autonomous AI techniques, offering automated discovery and delegated authorization at scale.
Automated playbooks reply immediately to identification anomalies. When malware triggers authentication irregularities, XSIAM revokes entry and launches forensic evaluation with out human intervention. This zero-latency response turns into the operational baseline.
Zscaler CEO Jay Chaudhry recognized the core vulnerability at Zenith Dwell 2025: “Community protocols had been designed to permit trusted units to speak freely. AI weaponizes this legacy structure at scale. Adversaries craft phishing campaigns that compromise agent identities quicker than people can reply.”
Common ZTNA frameworks allow million-agent deployments
The architectural necessities are clear. Common zero belief community entry (ZTNA) frameworks throughout the business present 4 capabilities important for AI environments.
Cisco’s implementation demonstrates the size required. Their Common ZTNA platform performs automated discovery scans each 60 seconds, cataloging new AI deployments and permission units. This eliminates blind spots that attackers goal. Cisco’s delegated authorization engine enforces least-privilege boundaries via coverage engines processing 100,000 choices per second.
Complete audit trails seize each agent motion for forensic investigation. Safety groups utilizing platforms like Cisco’s can reconstruct incidents throughout hundreds of thousands of interactions. Native help for requirements just like the Mannequin Context Protocol ensures interoperability because the ecosystem evolves.
Ivanti’s method enhances these capabilities with AI-powered analytics. Daren Goeson, SVP of product administration at Ivanti, emphasizes: “AI-powered endpoint safety instruments can analyze huge quantities of knowledge to detect anomalies and predict potential threats quicker and extra precisely than any human analyst. These instruments present clear visibility throughout units, customers and networks, proactively figuring out potential safety gaps.”
Cisco’s AI safety structure units business path
Cisco’s AI Safe Manufacturing facility positions them as the primary non-Nvidia silicon supplier in Nvidia’s reference structure. By combining post-quantum encryption with new units, Cisco is constructing infrastructure to guard towards threats that don’t but exist. The enterprise takeaway: Securing AI isn’t non-compulsory; it’s architectural.
At Cisco Dwell 2025, the corporate unveiled a complete identification and AI safety technique that addresses each layer of the stack:
| Announcement | Core downside solved / strategic worth | Technical particulars | Availability |
| Hybrid mesh firewall (incl. HyperShield) | Distributed, fabric-native safety; strikes safety from the perimeter into the community cloth | eBPF-based enforcement; {hardware} acceleration | New firewalls: Oct 2025 |
| Dwell shield | Closes “45-day patch vs. 3-day exploit” hole with speedy, kernel-level vulnerability shielding | Actual-time patching with out reboots | Nexus OS: Sept 2025 |
| Splunk: Free firewall log ingestion | Reduces SIEM prices as much as 80%; incentivizes Cisco firewall adoption | Limitless log ingestion from Cisco firewalls | Aug 2025 |
| Splunk: Observability for AI | Supplies vital visibility into AI stack efficiency | Screens GPU utilization and mannequin efficiency | Sept 2025 |
| Duo IAM | Evolves from MFA to a whole security-first IAM platform | Person Listing, SSO, Id Routing Engine | Accessible Now |
| Duo: Proximity verification | Delivers phishing-resistant authentication with out {hardware} tokens | BLE-based proximity, biometric verification | A part of the brand new Duo IAM |
| Duo: Id resilience | Addresses vital IDP outage dangers | Redundancy, load balancing and automatic failover | In growth |
| Cisco common ZTNA | Expands zero belief to people, IoT/OT units and AI brokers | Automated discovery, delegated authorization | Ongoing evolution |
| Open-sourced safety AI mannequin | Democratizes AI protection; 8B parameters match 70B mannequin efficiency | Runs on CPU; 5B safety tokens coaching | Accessible (Hugging Face) |
| AI protection and Nvidia partnership | Secures AI growth pipeline | Nvidia NIM microservices optimization | Accessible now |
| Submit-quantum safety | Future-proof towards quantum assaults | MACsec and IPsec encryption | New units (June 2025) |
| Id intelligence | Steady behavioral monitoring | AI-powered anomaly detection | A part of Safety Cloud |
| Safe entry | Converges VPN and ZTNA capabilities | Cloud-delivered safe entry service edge | Accessible now |
Cross-vendor collaboration accelerates
The Cloud Safety Alliance Zero Belief Development Heart now contains each main safety vendor. This unprecedented cooperation permits unified safety insurance policies throughout platforms.
“Safety distributors should unite towards frequent threats,” George Kurtz, CEO of CrowdStrike, emphasised throughout a current platform technique dialogue. “The info-centric method wins given how briskly adversaries and threats evolve.”
Cisco President and CPO Jeetu Patel echoed this sentiment in an interview with VentureBeat: “Safety is a prerequisite for adoption of AI. If individuals don’t belief the system, they’re not going to make use of it.”
The organizational problem stays. Robert Grazioli, CIO at Ivanti, identifies the vital barrier: “CISO and CIO alignment will likely be vital in 2025. This collaboration is important if we’re to safeguard trendy companies successfully. Executives have to consolidate sources — budgets, personnel, information and know-how — to boost a corporation’s safety posture.”
The identification reckoning
When Cisco, Okta, Zscaler, Palo Alto Networks and CrowdStrike independently attain an identical conclusions about identification structure, it’s affirmation, not coincidence.
Id infrastructure determines safety outcomes. Organizations face two choices: Architect identification because the management airplane or settle for breaches as inevitable. The hole between AI deployment velocity and identification safety maturity narrows every day.
Three actions can’t wait. Audit each AI agent’s identification and permissions inside 30 days. Deploy steady verification for all non-human identities instantly. Set up 24/7 identification safety operations to stop adversaries from exploiting gaps.
The seller consensus sends a transparent and unmistakable sign. Id has change into the management airplane for AI safety. Enterprises that fail to adapt will spend 2025 managing breaches as a substitute of innovation.
