A neon-blue futuristic cybersecurity and AI operations dashboard connects multiple laptops to analytics panels.
Every big endpoint vendor now sells a platform that goes beyond the endpoint, and the label "XDR" no longer tells you much about what you're buying. This piece compares CrowdStrike, Microsoft Defender and SentinelOne. It covers where each draws the line between EDR and XDR, how each prices its product, and how mature each one's AI-agent security is. It also sets out which published claims could be verified and which could not.

A neon-blue futuristic cybersecurity and AI operations dashboard connects multiple laptops to analytics panels. EDR and XDR: the practical difference​

EDR focuses on endpoint telemetry and response. Examples are investigating a device's activity and isolating the host. XDR connects endpoint signals with other sources such as identity, email, cloud and SaaS, correlates them and coordinates response across those domains. Which sources and actions you actually get depends on the licenses and integrations you deploy.

Microsoft's own documentation shows how blurry the line is. Defender for Endpoint is described as an enterprise endpoint security platform in the sense that its signals feed the unified Defender portal, where they are correlated with identity, email and cloud alerts into incident views. Microsoft still sells these as separate things. Defender for Endpoint is the EDR, Defender XDR is the correlation layer, and Microsoft Sentinel is a separate SIEM.

XDR is not a replacement for a SIEM. A SIEM stays vendor-neutral for log retention and custom detections. XDR trades some of that neutrality for pre-built correlation.

The buying decision is less about "EDR or XDR" and more about four questions:

  • Which data sources do you already have?
  • Which connectors does the vendor support?
  • What can be automated?
  • Can your team run one more console or service?

A team with a mature SIEM and identity stack may be best served by endpoint depth alone. A lean team may value a single incident queue, but only if the connectors and staffing support it.

Microsoft: bundling is the pricing model​

Microsoft's strongest card is packaging. Its documentation says Microsoft 365 E5 and Microsoft 365 E5 Security include Defender for Endpoint Plan 2. That is why an E5 customer's incremental cost for Defender EDR can be zero. Zero is not free, though. The cost sits inside a subscription that Microsoft's pricing page lists at $60.00 per user per month with Teams, or $51.45 without Teams, billed yearly. It also says prices may vary by agreement.

Other Microsoft pricing details worth knowing:

  • Defender Suite: Microsoft lists it at $12.00 per user per month for customers who already have Microsoft 365 E3. It positions the suite as the route to XDR across identities, endpoints, email and SaaS apps.
  • Vulnerability Management add-on: listed at $2.00 per user per month on an annual commitment.
  • XDR is not sold alone: Microsoft says Defender XDR is not a standalone offering. For standalone purchases it requires Defender for Endpoint Plan 2 and Defender for Office 365 Plan 2, and XDR features switch on automatically when several Defender workloads are in use.
  • Sentinel is separate: it is priced pay-as-you-go and needs an Azure subscription.
  • Device coverage: a Defender for Endpoint user license covers up to five devices per user, excluding servers, which are licensed separately.

Some third-party figures in circulation could not be confirmed in Microsoft's official material. These include a standalone Plan 2 price of $5.20 a month ($62.40 a year) and a $3.00 per user per month Defender for Business price. Treat them as reported figures and not as list prices. Check them against a current quote before they go into a budget.

CrowdStrike: per-device list prices​

CrowdStrike publishes its web prices per device. The Falcon Enterprise page lists $184.99 per device, billed annually, or $19.99 per device billed monthly. The same page shows Falcon Go at $7.99 monthly, Falcon Pro at $14.99 monthly and Falcon Enterprise at $19.99 monthly. Those are public web prices. A large enterprise should expect negotiated terms, and CrowdStrike's page does not describe enterprise-scale pricing.

On October 21, CrowdStrike announced an expanded collaboration with NVIDIA. Falcon Guardian, its AI Detection and Response product, is being integrated with NVIDIA OpenShell secure-runtime software. CrowdStrike said it plans an NVIDIA Sentry integration for in-silicon enforcement. That is a plan, not a shipped hardware-enforcement feature.

The original story also claims CrowdStrike extended agent coverage to Copilot Studio agents and Claude Code on July 30, 2026. I could not corroborate that date or scope in primary sources, so confirm it with CrowdStrike before building a requirement around it.

SentinelOne: a platform push with a consumption meter​

SentinelOne made two announcements at OneCon on October 21.

Singularity ADR. The company describes Singularity Agentic Detection and Response as a product that discovers AI agents, enforces what they may do, monitors behavior and stops unintended or malicious activity. Its press release says ADR is available in early access starting today. The vendor lists several capabilities:

  • An inventory of agents and MCP servers.
  • Coverage for Salesforce Agentforce, Microsoft Copilot Studio, OpenAI ChatGPT workspace agents and Anthropic Claude managed agents.
  • In-line blocking of prompt injection.
  • Containment that stops one agent mid-action without halting the whole workflow.

These are the vendor's claims. Early access is not general availability.

Singularity ASOC. SentinelOne also unveiled an Autonomous SOC suite. It combines data ingestion and analytics, agentic investigations and automated response, and includes AI SIEM, Hyperautomation, threat intelligence management and AI Data Pipelines with more than 50 third-party connectors. Coverage reporting adds two practical details:

  • ASOC draws down Singularity Credits, with administrator-set consumption guardrails.
  • Syslog and Kafka are supported today, while OpenTelemetry and generic API collection are listed as on the way.

I found no public SentinelOne list price. That is a limit of this research and does not prove none exists. For a consumption-metered product, ask for a written credit forecast before you sign.

PartnerOne also expanded the same day. LevelBlue, GuidePoint Security and World Wide Technology are launch partners, with a broader rollout planned for coming months.

AI-agent security: three different maturity levels​

The newest dividing line is how each vendor protects AI agents running on the same laptops as the EDR agent. The three products are at clearly different stages.

VendorProductStatusWhat is established
MicrosoftAI agent runtime protection in Defender for EndpointPreviewDetects prompt injection for local agents on Windows and can audit or block the action
SentinelOneSingularity ADREarly access (Oct 21, 2026)Vendor-described discovery, governance, in-line prevention and containment
CrowdStrikeFalcon Guardian (AIDR)NVIDIA OpenShell integration announced, Sentry plannedHardware enforcement is a plan, not a shipped feature

Microsoft's documentation is the most operationally specific. It lists these prerequisites:

  • Defender for Endpoint Plan 2, Microsoft 365 E5, Microsoft Agent 365 or Microsoft 365 E7.
  • Devices onboarded to Defender for Endpoint, with Defender Antivirus in active mode and real-time protection on.
  • A supported Windows version with current platform, engine and intelligence updates.
  • At least one supported local AI agent.

During public preview, Microsoft says to configure test devices to get platform and engine updates from the Beta Channel.

Rolling out Microsoft's preview safely​

Microsoft recommends a phased approach:

  1. Test: enable audit mode on a small set of devices where supported agents are actively used.
  2. Review: watch alerts in the Defender portal for one to two weeks and classify inaccurate ones as false positives.
  3. Deploy: extend audit mode to more device groups through Intune or the Defender portal.
  4. Enforce: switch to block mode only on groups where alert quality has been validated.

The settings take three modes: Disabled, Audit and Block. You can also set them locally in PowerShell, using the -AiAgentProtection and -AiAgentNetworkInspection parameters of Set-MpPreference.

In block mode, alerts carry a severity from Critical to Low. In audit mode, alerts are Informational. The same page says a "Suspicious AI prompt injection" alert appears on the device timeline and correlates into incidents.

The lesson applies to all three vendors: pilot a preview before you enforce it.

Comparing prices fairly​

A per-user monthly price and a per-device yearly price can't be compared until you fix a few assumptions. At minimum, settle these before comparing quotes:

  • Devices per user, remembering Microsoft's five-device allowance and the separate server licensing.
  • Server coverage.
  • Required add-ons, such as vulnerability management.
  • SIEM retention and data-ingestion charges.
  • MDR, if nobody covers nights and weekends.
  • Migration costs and the overlap period when two agents are licensed at once.
  • Any entitlements you already own.

A worked example shows why the unit matters. Take 100 users with one managed device each. CrowdStrike's published annual Enterprise price implies roughly $18,499 a year. A Microsoft estimate depends entirely on which plan you hold, so a standalone figure is meaningless for a company already on E5. That asymmetry is the real story. Microsoft is cheap at the margin for existing E5 customers, and CrowdStrike and SentinelOne are priced as replacements or additions.

What the analyst labels can and can't tell you​

The original story says all three vendors are Leaders in Gartner's 2026 Magic Quadrant for Endpoint Protection. I could not verify that placement in primary sources. CrowdStrike's own pricing page refers to the 2025 Gartner report, not 2026. Even if the 2026 placement is true, a shared Leader label doesn't separate the vendors.

No source I reviewed gives a verified false-positive rate or mean time to detect for all three on the same footing. Don't infer detection superiority from announcements or analyst badges.

A practical way to decide​

  1. Inventory first. List your operating systems, servers, existing agents, SIEM, identity provider and AI agents in use.
  2. Run a parallel pilot. Use the same representative devices and the same simulated activity for each candidate. Thirty days is a sensible window, though that figure is a reasonable practice and not a vendor requirement.
  3. Compare outcomes. Measure time to alert, investigation context, containment steps, analyst effort and connector reliability.
  4. Avoid dual real-time scanners. Don't leave two active real-time engines on one device during overlap.
  5. Start with detection-only modes. Raise automation ring by ring and don't enable automated containment everywhere on day one.

Bottom line​

The most defensible reading is a fit-based one. Microsoft is the natural default for Microsoft 365 E5 estates, because Plan 2 is already bundled. CrowdStrike offers transparent per-device list pricing and a platform story that extends toward SIEM and AI security. SentinelOne is making the widest new platform move, but ADR is in early access, ASOC is days old, and AI work is metered in credits. Preview labels change quickly, so recheck each vendor's status before finalizing a shortlist.

 

References

  1. EDR vs XDR: 3 Leaders, 3 Pricing Models (2026) shattered.io 2026-10-23T21:35:03+00:00
  2. CrowdStrike Falcon® Enterprise: Endpoint Breach Prevention crowdstrike.com
  3. SentinelOne Arms Partners to Build and Scale AI Security Practices | SentinelOne sentinelone.com