ARMOR In Action with ExtraHop: Securing the Agentic Enterprise from the Network

As AI evolves from pilot projects into business-critical systems, security becomes a defining factor in successful AI deployments. To help organizations address this challenge, WWT developed ARMOR (AI Readiness Model for Operational Resilience), a vendor-agnostic AI security framework. Built through collaboration with NVIDIA and strengthened by real-world engagement with the Texas A&M University System, ARMOR provides security leaders with a practical framework for assessing and managing AI risk. It organizes AI security into seven domains: Governance, Risk and Compliance, Model Protection, Secure AI Operations, Infrastructure Security, Identity Security, Data Protection, and Secure Software Development Lifecycle.

ARMOR's seven domains focus on key elements of securing AI systems and related processes, helping organizations understand security control points inside and between these systems, and their interactions with the surrounding environment. This is key to understanding how AI security mechanisms mesh with traditional security solutions and is vital to identifying present gaps. ARMOR also accommodates organizations at different maturity levels. ARMOR is a real-world-tested AI security framework that was shaped through two-way collaboration with a lighthouse customer, Texas A&M University, who served as the first testing ground.

ExtraHop brings a distinctive perspective to the ARMOR framework. Most AI security controls sit inline and at the perimeter: a gateway, a firewall, an agent on a host. ExtraHop RevealX observes from the network itself, passively and out of band, at line rate, within the network (east/west) and at the perimeter (north/south). Every LLM call, MCP tool invocation, agent-to-agent exchange, Kubernetes service call, and authentication handshake crosses the wire, and RevealX decrypts and decodes that traffic in real time across more than 90 protocols. The result is a live inventory of what is actually running, what it talks to, and whether that behavior matches baseline.

ExtraHop was named a Leader in the Gartner Magic Quadrant for Network Detection and Response in 2025 and again in 2026. RevealX Federal holds FedRAMP Moderate authorization. In August 2026 the company announced the first NDR sensor to analyze traffic in full at 400 Gbps, enabling compatibility with NVIDIA AI factory fabrics such as Spectrum-X Ethernet. WWT customers can evaluate this alongside endpoint telemetry in the Advanced Technology Center's CrowdStrike and ExtraHop Integrated Lab. In the sections ahead, we'll show how ExtraHop's solutions map to the ARMOR domains, with particular strength in Secure AI Operations, Infrastructure Security, and the runtime side of Model Protection, reinforced by identity-to-network correlation that few platforms can match.

Mapping ExtraHop's portfolio to ARMOR domains

ExtraHop's portfolio is one platform, RevealX, delivered as RevealX 360 (SaaS-managed) or RevealX Enterprise (self-managed) on a common sensor and analytics engine. Capabilities are packaged as modules (NDR, NPM, IDS, Packet Forensics) and as solution areas built on the same wire data: AI estate visibility, identity threat investigations, Kubernetes and cloud workload visibility, encrypted traffic analysis, and an open context layer (REST API, EQL, and MCP server) for the agentic SOC.

Because every capability draws on the same evidence, coverage in one ARMOR domain reinforces the others. The mapping also lines up with ARMOR's own sub-topics: Risk content and Shadow AI and Agentic architecture governance under GRC; Threat detection and the SOC of the Future under Secure AI Operations; Network, Host and DPU Security and Quantum under Infrastructure Security; Agentic Identities under Identity Security; and DLP under Data Security. ExtraHop's strongest, most-cited coverage sits in Secure AI Operations (RevealX NDR, the agentic SOC context layer, and the integration ecosystem), Infrastructure Security (line-rate decryption, Kubernetes and cloud visibility, IDS, and the 400 Gbps sensor), and Model Protection at runtime (AI asset inventory, LLM and MCP observability, and AI-specific detections). Identity Security and GRC are strong overlaps anchored by identity correlation and immutable packet evidence. Data Security is scoped to detecting sensitive data movement rather than classifying or encrypting data at rest, and the Secure Software Development Lifecycle is supported as runtime validation of what development teams ship rather than as code scanning.

Across the seven ARMOR domains, ExtraHop's alignment can be summarized as follows:

With the portfolio mapped, the following sections describe how each ExtraHop solution contributes to the ARMOR domains it primarily supports and the domains it secondarily reinforces.

RevealX NDR

ARMOR Domains: Secure AI Operations (Primary); GRC, Model Protection, Infrastructure Security (Overlap)

AI attacks now move at machine speed. ExtraHop's August 2026 announcement cites a mean time to exploit of roughly 1.6 days in 2026, down from 23.2 days a year earlier. Secure AI Operations at that pace cannot depend on a log being forwarded, or a traffic sample being representative. RevealX NDR analyzes full packet streams in memory, reconstructs flows and L7 transactions, and runs rule-based, machine learning, and retrospective detections against them in parallel. Notable capabilities include:

• Behavioral detection across 90+ protocols: Decodes the Microsoft protocols attackers use for lateral movement (NTLM, Kerberos, LDAP, MS-RPC, SMB, WSMAN) and baselines normal behavior per device, service, and identity.
• Three detection sources, one workflow: On-box rules, cloud-scale ML, and Automatic Retrospective Detections land in the same investigation, each mapped to MITRE ATT&CK, with AI-specific detections aligned to MITRE ATLAS techniques.
• Continuous monitoring for NIST AI RMF and ISO/IEC 42001: Supplies the runtime measurement that the AI RMF Measure and Manage functions and the ISO/IEC 42001 monitoring clauses call for, drawn from observed behavior rather than self-reported controls.
• Threat briefings and intelligence: Cloud-delivered briefings for emerging campaigns, plus curated intelligence from CrowdStrike Falcon Adversary Intelligence, Hunt.io, and MISP feeds.
• Out-of-band: Sensors receive a copy of traffic from a tap, SPAN, packet broker, cloud mirror, or software traffic forwarder
• Investigation and response in one console: Guided pivots from detection to device, identity, records, and packets, with in-platform quarantine, session revocation, and user disablement through integrations.

AI estate visibility

ARMOR Domains: Model Protection (Primary); GRC, Data Security, Secure AI Operations (Overlap)

You cannot protect a model you did not know was deployed. ARMOR files Shadow AI and Agentic architecture governance under GRC and makes Shadow AI a featured cross-domain use case, and this is where ExtraHop's AI estate capabilities land first. ExtraHop's 2026 Global Threat Landscape Report found that 55% of security leaders name AI agents and GenAI applications their top attack-surface risk, and 35% of incidents trace back to employees feeding proprietary data into unvetted public AI tools. In March 2026 ExtraHop announced capabilities that discover AI infrastructure the moment it starts communicating, then observe what it does. This is Model Protection delivered as runtime observation and detection, and it complements inline gateways and guardrails by verifying from the network that they are in the path and working. Core capabilities include:

• AI asset inventory: Automatically discovers LLM usage, MCP servers, APIs and tool endpoints, and agents with their communication patterns, so approved AI can be baselined and unsanctioned AI flagged as soon as it appears.
• LLM request and response observability: Decrypts and parses LLM traffic in real time to spot anomalies in intent, prompt structure, or data output, and to record which model, which user, and what data was involved.
• MCP-mediated tool visibility: Surfaces which internal systems an agent calls through MCP and the actions it performs there, exposing high-privilege gateways to sensitive data.
• AI governance evidence: Detects bypass of approved AI gateways, use of unsanctioned models, and non-compliant data flows, with an audit trail of agent activity for policy enforcement.
• Agentic identity propagation: Follows how credentials and permissions move across multi-step agentic workflows, so token reuse by a non-human identity is visible as a chain rather than isolated events, supporting ARMOR's Agentic Identities guidance.

Identity threat investigations

ARMOR Domains: Identity Security (Primary); Secure AI Operations (Overlap)

Attackers increasingly log in rather than break in, and AI multiplies the non-human identities in play: service accounts, workload identities, API keys, and agent credentials. ARMOR's Identity Security domain frames this as Identity and Zero Trust for the agentic era, and Forrester's observation that there can be no Zero Trust without visibility into what is happening inside the network applies directly. In February 2026 ExtraHop added integrations with Microsoft Entra ID, Active Directory, and Okta that fuse identity attributes with network telemetry in a single data set. The alignment to Identity Security is deliberately scoped: RevealX detects and investigates identity abuse and provides direct containment, complementing rather than replacing lifecycle controls such as MFA, vaulting, and federation. Highlighted capabilities include:

• Identity as a first-class object: Users are searchable and pivotable like devices, enriched with title, department, and manager, so analysts and agents see who acted alongside what the network shows they did.
• Risky user detections and dashboards: Entra ID dashboards highlight unusual user and administrative activity; detections cover impossible travel, password spray, anomalous token use, and forged or suspicious access.
• Decrypted authentication protocols: Passive decryption of Kerberos, NTLM, LDAP, MS-RPC, SMB 3, and WSMAN exposes credential abuse and remote PowerShell commands run over encrypted channels.
• Blast radius from the wire: Confirms lateral movement and scopes a compromised account from observed network activity, not only from identity provider logs an attacker may have evaded.
• In-platform response: Revoke sessions or disable users directly from the investigation.

Encrypted traffic analysis

ARMOR Domains: Infrastructure Security (Primary); Model Protection, Data Security (Overlap)

ARMOR's Infrastructure Security domain spans Network, Host and DPU Security through to Quantum, and the network layer is where ExtraHop contributes. Agent-to-agent traffic, LLM API calls, and service-to-service communication inside Kubernetes are almost entirely encrypted. In the first verified AI-orchestrated espionage campaign, the adversary's agent ran most of the intrusion inside encrypted sessions that looked identical to routine API calls. ExtraHop's position is that encryption hides payloads, not behavior, and where keys are available it does not have to hide payloads either. Capabilities include:

• Out-of-band TLS decryption including TLS 1.3: Out-of-band decryption using forwarded session keys or integrations with F5 LTM, NetScaler, Netskope, and Zscaler, with no inline device and no added latency.
• Microsoft protocol decryption: Makes the protocols attackers rely on for east-west movement readable, not just counted.
• mTLS visibility inside Kubernetes: Decrypts service-to-service communication in clusters where container escapes and hijacked workload credentials otherwise operate unseen.
• JA4+ fingerprinting when decryption is not an option: The first NDR to ship the full NDR-addressable JA4+ suite built in, identifying client software, C2 frameworks, and anomalous TLS behavior from handshake characteristics.
• Quantum Readiness evidence: Records the key agreement negotiated per session (RSA key size, DHE group, ECDHE curve, or hybrid ML-KEM), so the ARMOR Quantum Readiness use case has measurable data on PQC migration progress and harvest-now-decrypt-later exposure.

Kubernetes and cloud workload visibility

ARMOR Domains: Infrastructure Security (Primary); Secure AI Operations, Secure Software Development Lifecycle (Overlap)

Production AI increasingly runs as containers on Kubernetes across hybrid and multi-cloud infrastructure, and containers are ephemeral by design. RevealX treats pods and nodes as discovered devices with their Kubernetes metadata attached and preserves a history of record, so all the behaviors performed by a workload  outlives the workload. Even though Kubernetes adds a lot of concepts, what hasn't changed is that operators often are solely reliant upon self-reported workload logs – which can be inaccurate and incomplete. ExtraHop provides an independent source of truth based on wire data, which is just as relevant in Kubernetes environments as it is anywhere else. Capabilities include:

• Pods and nodes as assets: Every pod and node appears with relevant Kubernetes properties like namespace, cluster, deployment, and ReplicaSet, giving security and platform teams one view of what is running.
• Correlated container detections: Ties Kubernetes behavior to identity data (a container acting on a hijacked credential) and to threat intelligence (a third-party image talking to known C2 infrastructure).
• Hybrid and multi-cloud coverage: Sensors deploy in AWS, Azure, and Google Cloud through cloud packet mirroring or packet forwarders, Flow Sensors ingest VPC Flow Logs and NetFlow, and provisioning is automatable with Terraform and Ansible.
• Runtime validation for the SDLC: When a team ships a new agent or LLM integration through software-orchestrated infrastructure, RevealX shows its real dependency map, latency, and traffic pattern before a security review or outage forces the conversation.

Packet Forensics

ARMOR Domains: GRC (Primary); Secure AI Operations, Data Security (Overlap)

Regulators, insurers, and courts ask organizations to prove what happened, not describe what they think happened. Logs can be spotty, incomplete, or misleading; packets captured out of band cannot. The Packet Forensics module stores full packets in a Packetstore on an independent feed, while the Recordstore keeps L7 transaction records (individual commands and responses per protocol) searchable for 30 to 365 or more days. Highlighted capabilities include:

• Immutable audit trail: Evidence the monitored system never knew was captured cannot be tampered with from a compromised asset, meeting tamper-proof expectations in DORA, FCA PS21/3, and NYDFS Part 500, and providing the documented evidence NIST AI RMF and ISO/IEC 42001 audits ask for.
• Data movement evidence for DLP: Transaction records show what data crossed which AI interaction and to which destination, giving the Data Security domain's DLP controls an independent record to validate against.
• Transaction-level records: Faceted search answers which user called which model with what data, or which MCP tool an agent invoked at a specific timestamp.
• Retrospective detection: Stored records are rescanned when new indicators are published, surfacing historical compromise that was unknown when it occurred.
• Long-term retention options: Records export to CrowdStrike LogScale, Google BigQuery, and Amazon Security Lake for legal, insurance, and regulatory workflows.

IDS module

ARMOR Domains: Infrastructure Security (Primary); Secure AI Operations (Overlap)

Behavioral detection finds the unknown; signatures find the known fast. The IDS module runs on the same sensor as NDR, so AI infrastructure gets both without a second appliance or a second copy of traffic. Capabilities include:

• Curated signature coverage: Vendor-managed rules including Proofpoint ET Pro, with support for customer-provided signatures.
• In-product rule catalog: Transparency into the active ruleset and its CVE coverage for audit and compliance teams.
• Signature plus context: IDS matches land in the same investigation as behavioral detections, where records and packets confirm whether a match reflects real exploitation.

RevealX NPM

ARMOR Domains: Infrastructure Security (Primary); Secure AI Operations, Secure Software Development Lifecycle (Overlap)

ARMOR treats cyber resilience as a through-line, and for AI that means the availability and performance of GPU clusters, inference endpoints, and the shared backbones agents depend on. RevealX NPM uses the same wire data as NDR to give operations and platform teams dependency maps and performance telemetry for infrastructure that cannot be instrumented. Notable capabilities include:

• AI backbone monitoring: Out-of-band monitoring of 100 Gbps and, from Q4 2026, 400 Gbps fabrics, including NVIDIA-based AI factory networks, with no added latency, exposing infrastructure strain and runaway consumption as it happens.
• Dependency and activity maps: Real service-to-service maps for the systems an agent or model calls, so an integration that overloads a shared dependency is visible before users complain.
• One ground truth for SOC and NOC: Security and operations teams, and their agents, reason over the same view, removing the device-count disputes fragmented tools create.

Agentic SOC context layer

ARMOR Domains: Secure AI Operations (Primary); GRC, Secure Software Development Lifecycle (Overlap)

ARMOR describes Secure AI Operations as the SOC of the Future, where reactive defenses become proactive strategies. An autonomous agent that reasons from partial evidence can be confidently wrong at scale, and its audit trail will look sound right up to the breach nobody can explain. ExtraHop publishes network context in a form agents query directly: structured, discoverable, with relationships intact. In July 2026 ExtraHop launched the Agentic SOC Alliance with founding members including CrowdStrike, Torq, Dropzone AI, Prophet Security, Exaforce, Intezer, Command Zero, LangChain, Kindo, AuthMind, Armadin, ReversingLabs, Fig, and TENEX.AI, to standardize a three-layer operating model of Context, Harness, and Model. RevealX is the Context layer. Highlighted capabilities include:

• Open MCP server and agent skills: Published on GitHub, so customer-built or partner agents query the same surfaces ExtraHop's own tooling uses, with no proprietary runtime.
• ExtraHop Query Language (EQL): Lets agents and analysts pull exactly the context an investigation needs from very large volumes of telemetry at machine speed.
• Layered evidence on demand: Agents start with detections, assets, and relationships and drill into metrics, L7 records, and packets only when a case requires it, lowering token consumption, latency, and cost per investigation.
• Bidirectional REST API and streaming: Read and write access to detections, devices, records, and configuration, with open data streams to SIEM, data lakes, and SOAR.
• Governed automation: Detection tuning, RBAC, and audit logging on the platform, with human approval routing handled in the harness where policy requires it.

Integration ecosystem

ARMOR Domains: Secure AI Operations (Primary); Infrastructure Security, Identity Security (Overlap)

No single vendor delivers autonomous defense alone. ExtraHop's integrations push network context into the tools teams already run and pull response actions back out, either built in (click the partner tile) or customized through the REST API and triggers. Capabilities include:

• SIEM and data platforms: CrowdStrike Falcon Next-Gen SIEM and LogScale, Splunk, Microsoft Sentinel, Google Security Operations, IBM QRadar, Exabeam, Sumo Logic, Datadog, and Amazon Security Lake.
• EDR and XDR correlation: CrowdStrike Falcon (including discovery of endpoints with no Falcon agent), Microsoft Defender for Endpoint and XDR, SentinelOne, Trellix, Symantec, and Carbon Black, with automated quarantine.
• Network enforcement and zero trust: Palo Alto Networks Panorama and Cortex XSOAR, Cisco ISE and Meraki, Fortinet, Check Point, Sophos, Forescout, Aruba ClearPass, Zscaler ZPA, and Netskope.
• SOAR, ITSM, and collaboration: Splunk SOAR, Tines, Torq, ServiceNow ITSM and Service Graph Connector, Jira, PagerDuty, Slack, and Microsoft Teams.

400 Gbps sensor

ARMOR Domains: Infrastructure Security (Primary); Secure AI Operations (Overlap)

When data centers and neoclouds moved to 400 Gbps, the speed NVIDIA Spectrum-X™ Ethernet and comparable AI fabrics are built for, most detection tools stayed at 100 Gbps and began sampling or clustering sensors at significant cost. GPU clusters, training and inference pipelines, and agent-to-agent traffic generate dense, encrypted east-west volumes where sampling means blind spots. Announced in August 2026 and generally available in Q4 2026, ExtraHop's 400 Gbps sensor analyzes traffic in full at line rate. Capabilities include:

• Complete evidence, not a sample: An agent's conclusions rest on what actually happened, including the encrypted east-west traffic where modern intrusions live.
• Structured context at scale: Traffic becomes a live semantic map of every device, identity, workload, and conversation, queryable through API and MCP.
• Lower total cost: Higher per-sensor throughput reduces sensor count, clustering overhead, and operational complexity.

Operationalizing ExtraHop across ARMOR

One evidence source, seven domains

ExtraHop's approach rests on a single premise: the network is the one control plane an attacker cannot silence. Endpoint agents can be uninstalled, logs can be disabled, and application-layer controls can be bypassed by an agent that was never registered with them. Traffic still has to cross the wire. RevealX reuses that one evidence source across all seven domains, producing the AI inventory GRC needs, the runtime detections Model Protection needs, the east-west visibility Infrastructure Security needs, the identity correlation Identity Security needs, the sensitive-data-flow detection Data Security needs, the production validation the SDLC needs, and the structured context Secure AI Operations needs. Every conclusion in any domain traces back to packets.

How the pieces fit together

Inside the portfolio, integration is architectural rather than bolted on. Metrics, records, packets, detections, identities, and Kubernetes metadata come from the same sensors and are federated by the same console, so an analyst or agent moves from an AI governance violation to the identity behind it, to the L7 record of the exact prompt, to the packet, without switching tools. With third parties, ExtraHop plays the Context role in the Alliance's Context, Harness, and Model architecture: RevealX supplies evidence, a harness such as Torq, Kindo, Splunk SOAR, or Cortex XSOAR governs what agents may do with it, and the model is interchangeable. Enforcement flows outward to EDR, firewalls, NAC, SSE, and identity providers.

WWT recommends the following practices for maximizing value across the portfolio:

• Instrument east-west first: Place sensors where AI workloads talk to each other and to data: GPU fabrics, Kubernetes ingress and service mesh, and the segments hosting MCP servers and vector databases.
• Feed session keys early: Deploy key forwarding on LLM gateways, load balancers, and Kubernetes ingress, and use JA4+ where keys cannot be obtained.
• Connect identity on day one: Enable the Entra ID, Active Directory, or Okta integration before the first investigation so every detection carries a who.
• Register sanctioned AI: Baseline approved models, MCP servers, and agents so unsanctioned entities are flagged against a known-good register.
• Size retention to the regulator: Set Recordstore and Packetstore retention to the longest applicable evidence requirement rather than default storage.
• Expose context through MCP, not dashboards: Point existing AI SOC platforms at the ExtraHop MCP server and EQL so they query structured context instead of raw logs.
• Close the loop with enforcement: Pair high-confidence detections with automated containment, and keep human approval in the harness for actions that touch production AI services.
• Give the NOC the same console: License NPM alongside NDR so platform and security teams work from one data set.

Use cases and industry adaptations

Regulated industries. Financial firms under DORA, FCA PS21/3, NYDFS Part 500, APRA CPG 230, and MAS guidelines must identify critical services, map dependencies, monitor continuously, and prove what happened during an incident. RevealX delivers all four from immutable wire data, and adds post-quantum readiness reporting on which TLS and SSH services still negotiate quantum-vulnerable key exchange. Fiserv's CISO, speaking at the Agentic SOC Alliance launch, described real-time network and endpoint telemetry as the primary substrate for the firm's AI agents, with the SIEM kept for depth rather than speed.

High-performance AI. Neoclouds and enterprise AI factories on 400 Gbps fabrics cannot tolerate inline inspection latency or sampled visibility. Out-of-band sensors at line rate, with Kubernetes and mTLS decryption, give these environments full east-west coverage without touching the GPU data path. WWT's Chris Konrad, Vice President of Global Cyber, put the requirement plainly in ExtraHop's 400 Gbps announcement: AI-powered defenses need to tell the difference between a quiet network and one they are only partially seeing.

Hybrid cloud. RevealX 360 manages sensors across on-premises data centers, AWS, Azure, and Google Cloud from one SaaS console, and the Zscaler and Netskope integrations extend visibility into SSE tunnels for remote users calling AI services. The Gigamon, ExtraHop, and WWT Power of 3 offering packages traffic access, analysis, and services for these deployments.

Healthcare. Medical devices, imaging systems, and legacy clinical applications cannot run agents, so passive network visibility is the only way to inventory and monitor them alongside the AI models now reading their output. ExtraHop reports securing more than 50 million patient records, with deployments emphasizing HL7 and DICOM decoding, ransomware precursor detection on SMB, and immutable evidence for breach notification decisions.

Manufacturing and critical infrastructure. OT and IoT environments are agentless by nature and increasingly connected to AI-driven analytics. RevealX discovers and profiles unmanaged devices, decodes industrial protocols, and detects lateral movement between IT and OT segments without changing the OT network.

Retail and public sector. ExtraHop reports more than 2 million point-of-sale systems under monitoring; Home Depot used RevealX to improve application performance 56% and lift buy-online-pick-up-in-store revenue 40%. RevealX Federal's FedRAMP Moderate authorization (October 2025) lets federal civilian and defense agencies deploy NDR for AI workloads within existing authorization boundaries.

Metrics and KPIs for CISOs

WWT recommends measuring ExtraHop's contribution to ARMOR-aligned governance, per solution area and across the portfolio, with the following indicators:

• AI asset coverage ratio: Discovered LLM endpoints, MCP servers, and agents versus the sanctioned AI register; the gap is your shadow AI exposure.
• Time to sanction or remove: Median hours from RevealX flagging an unsanctioned AI service to a governance decision.
• Encrypted visibility percentage: Share of east-west and AI-related traffic that is decrypted or JA4+-fingerprinted rather than opaque.
• Throughput coverage: Percentage of data center and AI fabric traffic analyzed in full versus sampled or unmonitored.
• Mean time to detect and respond: Tracked per detection category (AI-specific, identity, lateral movement), against ExtraHop-published customer outcomes of roughly 83% faster detection and 87% faster response.
• Dwell time: Days between first observed attacker activity, including retrospective detections, and containment.
• Identity attribution rate: Percentage of detections and investigations that carry a resolved user or workload identity.
• Evidence-backed closure rate: Percentage of investigations closed with record- or packet-level evidence, and median time to retrieve that evidence for audit or legal requests.
• Retention compliance: Days of searchable record and packet history versus the longest applicable regulatory requirement.
• Automated containment rate: Percentage of high-confidence detections that trigger an integrated response without manual intervention, alongside the false-positive rate of those actions.
• Agent investigation efficiency: Tokens, latency, and cost per autonomous investigation when agents query structured ExtraHop context versus raw telemetry.
• Portfolio consolidation: Point tools retired (IDS, packet capture, NPM, flow analysis) and analyst hours recovered.

How customers benefit

A customer building on ExtraHop gains one source of truth that serves the CISO, the CIO, and the AI platform team at once. Security gets an inventory of the AI estate it did not have, detections for AI-specific and identity-driven attacks that other tools cannot see inside encryption, and evidence that stands up to a regulator. Operations gets performance and dependency data for GPU clusters and agent workloads with zero added latency. The AI team gets early warning when a new agent misbehaves in production. Any agentic SOC platform the customer already runs gets better, because the context it reasons over is complete, structured, and current. ExtraHop-published customer outcomes include roughly 83% faster detection, 87% faster response, 92% faster outage resolution, and a three-year ROI near 193%.

Innovation and roadmap

ExtraHop's 2026 roadmap follows the threat. The 400 Gbps sensor reaches general availability in Q4 2026. The Entra ID, Active Directory, and Okta integrations, in public preview since February 2026, are moving toward general availability, and EQL is moving from beta to the full security stack. The Agentic SOC Alliance is producing requirements and implementation blueprints for the Context, Harness, and Model architecture, with ExtraHop publishing evidence semantics so any member's agent can discover and query network context; ExtraHop-built agents remain a roadmap direction. The AI estate capabilities announced in March 2026 continue to expand protocol coverage for LLM APIs, MCP, and agent frameworks as those protocols evolve. On the regulatory side, post-quantum key agreement visibility already provides measurable evidence for NIST ML-KEM migration, detections track the OWASP Top 10 for LLM and MITRE ATLAS as those catalogs evolve, and the immutable evidence model supports NIST AI RMF, ISO/IEC 42001, and operational resilience regimes as they extend into AI systems. ExtraHop stays ARMOR-aligned by design: because coverage derives from the network rather than from a specific model, framework, or vendor, a new agent protocol or evidence requirement becomes a decoder or a report on the same platform, not a new product.

Where ExtraHop is uniquely strong

ExtraHop's portfolio demonstrably leads in three ARMOR domains.

Secure AI Operations. ExtraHop is a two-time Gartner Magic Quadrant Leader for NDR, founded the Agentic SOC Alliance, and publishes an open MCP server and agent skills so autonomous SOC platforms query structured network context directly. RevealX produces evidence while an attack is unfolding, at line rate, and hands it to agents pre-correlated across assets, identities, behaviors, and threats, which lowers both error rate and cost per investigation compared with reasoning over raw logs.

Infrastructure Security. ExtraHop is the first NDR to analyze traffic in full at 400 Gbps, where alternatives sample or cluster. Combined with out-of-band TLS 1.3 and Microsoft protocol decryption, mTLS visibility inside Kubernetes, the first built-in full JA4+ suite, and coverage of GPU nodes, OT, IoT, and legacy systems, ExtraHop sees the east-west traffic where AI-era intrusions actually move, with zero impact on the data path.

Model Protection at runtime. ExtraHop observes the model layer from the network rather than from inside the application. It inventories every LLM, MCP server, and agent the moment it communicates, reads decrypted prompt and response traffic, follows identity propagation across agentic workflows, and detects anomalous agent behavior as network patterns. That makes it the independent check on whether inline gateways, guardrails, and AI firewalls are actually in the path and actually working.

Identity Security deserves a fourth call-out as the differentiated overlap: identity explains who, the network explains what actually happened, and RevealX holds both in the same investigation.

Ready to get started?

At WWT, we're helping organizations operationalize ARMOR with the partners who can deliver against it. ExtraHop's portfolio brings particular strength in Secure AI Operations, Infrastructure Security, and runtime Model Protection; anchored by RevealX NDR and the open agentic SOC context layer for machine-speed operations, line-rate decryption and the 400 Gbps sensor for the infrastructure layer, and AI estate visibility with identity correlation for the models, agents, and credentials that make up the new attack surface.

Whether you're standing up a new AI inference platform or bolstering the security posture of an existing one, the path forward starts with visibility and prioritization. WWT's Security practice pairs an ARMOR assessment with hands-on RevealX expertise in the Advanced Technology Center, where the CrowdStrike and ExtraHop Integrated Lab lets teams see network and endpoint evidence working together before deployment, then carries through sensor placement, decryption design, identity integration, and agentic SOC enablement.