Ten years ago, corporate protection was built around signatures: an antivirus compared a file with a database of known malicious samples and blocked any match. Today, that approach is losing. Attackers generate thousands of malware variants every day, abuse legitimate system tools (PowerShell, WMI, remote administration software) and increasingly operate without files at all. That is why leading vendors have bet on artificial intelligence that asks not “what is this file?” but “what is happening in the system?”.
The two platforms that appear most often in corporate tenders are CrowdStrike Falcon and Microsoft Defender. Both belong to the EDR/XDR class (Endpoint/Extended Detection and Response), but they use machine learning in different ways. Let’s look at how.
Why classic antivirus is no longer enough
A modern attack rarely looks like a single “virus”. It is a chain of actions: a phishing email, stolen credentials, a login under a legitimate user, privilege escalation, lateral movement across the network and only at the end, encryption or data theft. Each individual step may look normal. The anomaly becomes visible only when the system sees the entire sequence of events and understands the context.
This is where AI has an advantage over people and static rules: models can correlate millions of events from different devices, cloud services and accounts in real time and spot combinations that match attacker tactics.
CrowdStrike Falcon: cloud intelligence and behavioural indicators
Falcon is built as a cloud platform with a single lightweight agent (sensor) on every device. The sensor streams telemetry to the cloud, where it is processed by the Threat Graph, a database that analyses trillions of events from customers worldwide every week. The platform’s key AI mechanisms are:
- Machine learning on the device and in the cloud. Models assess files and processes before they run, so they block even new, never-seen-before samples without signature updates.
- Indicators of Attack (IOA). Instead of looking for traces of known malware, Falcon tracks behaviour: for example, an office document that launches a command shell and reads the memory of the LSASS process is a typical credential-theft scenario.
- Charlotte AI. A generative AI assistant that analysts can ask questions in natural language (“which hosts contacted this domain last week?”) and receive an answer together with a ready-made query.
- Falcon OverWatch. A 24/7 threat-hunting team that complements the algorithms with human expertise and reviews complex cases.
Microsoft Defender: the ecosystem and automatic response
Microsoft’s strength is integration. Defender for Endpoint is built into Windows, while Microsoft Defender XDR combines signals from devices, email (Defender for Office 365), identities (Microsoft Entra ID) and cloud apps. Microsoft’s models are trained on trillions of signals the company receives every day from its global infrastructure.
- Automated investigation and response (AIR). After an alert, the system investigates related files, processes and accounts on its own and then removes malicious artefacts, often before an analyst even opens the incident.
- Automatic attack disruption. When XDR sees an attack such as ransomware or business email compromise with high confidence, it automatically contains the device or disables the compromised account to stop the spread.
- Microsoft Security Copilot. A generative assistant that summarises incidents, explains attacker scripts and helps write threat-hunting queries.
Comparing the approaches
| Criterion | CrowdStrike Falcon | Microsoft Defender |
|---|---|---|
| Architecture | Cloud platform, one agent for multiple operating systems | Built into Windows, deep integration with Microsoft 365 and Azure |
| AI strength | Behavioural IOAs, global Threat Graph | Correlation of email, identity and device signals |
| Automation | Host containment, Falcon Fusion response workflows | AIR and automatic attack disruption |
| Best fit | Multi-platform environments, companies at high risk of targeted attacks | Organisations already working in the Microsoft ecosystem |
How to roll out AI protection without chaos
- Inventory. Identify every endpoint, server and cloud workload: AI cannot protect a device that has no agent.
- Pilot in monitoring mode. For the first few weeks, run the platform in detection mode to assess false positives and tune exclusions.
- Gradual automation. Automate the obvious actions first (containing a clearly malicious host) and leave complex decisions to analysts.
- Update control. The 2024 incident, when a faulty update from a single vendor took down millions of Windows computers, showed that security tool updates should also be rolled out in stages.
- Team training. AI assistants speed up the work, but conclusions and accountability remain with people.
Artificial intelligence does not replace basic hygiene: updates, multi-factor authentication and backups. But it is AI that makes it possible to spot an attack in minutes rather than weeks, and that often decides whether an incident becomes a disaster.