The security tool
for the AI-powered malware
of tomorrow.

AI Malware Guardian is designed to protect against newly emerging AI-powered malware. These threats are incredibly evasive, and traditional antivirus tools struggle to defend against them.

Our approach solves the problem by using behavioral and structural baselines instead of signature database libraries. Shipped ONNX models run on your PC; you record personal baselines so detection compares activity to what normal looks like on your machine—not a global signature list.

Take a step towards defending against the future of malware.

Download for Windows See How It Works
Local-first detection · Guided response workflow · Complete Data Privacy · Dual ML Models · Real-Time Process Monitoring

The threat is already changing

AI doesn't just power security tools.
It is now embedded into the malware attacking you.

It rewrites itself to stay hidden

AI-powered malware can mutate its own code between executions, generating variants that look completely different to a signature scanner each time. A threat your antivirus blocked yesterday can return today in a form it has never seen.

It studies your system before it strikes

Modern AI malware performs reconnaissance first, profiling your installed software, security tooling, and usage patterns before deciding when and how to deploy a customized payload. It is designed to wait for the right moment to go unnoticed.

The payloads are severe

Once inside, AI-assisted malware has been used to deploy ransomware, exfiltrate credentials, establish persistent backdoors, and enroll machines into botnets while generating decoy activity to mislead forensic analysis. The consequences are no longer theoretical.

Our golden question

How can we defend against malware that has never been seen before?

Every traditional antivirus works from a list of known threats, signatures, and patterns. If malware is brand new or able to morph its own code to avoid detection, that list is useless. Guardian takes a completely different approach: instead of asking "have I seen this before?" it asks "does this look normal for this machine?" It builds a precise picture of what your computer normally does, and anything that deviates from that picture is flagged. This theoretical approach protects against the future and the unknown, regardless of how advanced the malware becomes.

Built different from traditional AV

Traditional antivirus reacts to known threats. AI Malware Guardian detects unknown ones.

Local-first detection

Core scoring runs on your hardware with shipped ONNX models and baselines you record—your files, processes, and behavioral features are not uploaded for cloud inference. Expected outbound calls are limited to optional signed app updates and your website account when you download the installer—not continuous behavioral telemetry.

Personalized to Your Machine

Most security tools compare you against a global average. Guardian is unique because it learns what normal looks like specifically on your machine, shaped by your hardware, your software stack, and your day-to-day patterns. You are encouraged to record personal baselines for your system and specific applications to improve detection on your PC. Your protection is unique to you because your machine is unique.

Signatures Go Stale. Anomalies Don't.

Traditional antivirus writes its rules after a threat is already found. By definition, it can't catch what's new. Guardian doesn't need to know a threat by name. It just needs to see behavior that strays from your baseline. A novel AI-generated variant with no signature, no hash, no name in any database still deviates from known-good. That deviation is the detection.

Real-Time Behavioral Detection

Watches Windows kernel event traces (ETW) and flags processes whose runtime behavior diverges from the benign baseline, catching even brand-new, never-seen-before malware. Every flagged process is assessed against its origin path and known application signatures before Respond suggests next steps. Guardian does not move or delete files for you.

ETW Fast-Path Monitoring

Guardian watches Windows kernel event traces (ETW) and scores short per-process event windows with Layer 1a. Unusual bursts are flagged while activity is happening, without sending behavioral data to the cloud.

Respond when action is required

Tier 4 alerts open a step-by-step Respond guide: review evidence, check path reputation, and hand off to Windows Security or your own workflow. Nothing is quarantined automatically in the open-source edition.

Learns what's normal. Flags everything that isn't.

AI Malware Guardian combines ETW fast-path scoring (Layer 1a) with rolling machine-wide behavioral monitoring (Layer 1b). Both ONNX models run on-device and feed the same alert pipeline.

Baseline

Build Your Baseline

Guardian's behavioral detection only works because it first learns what normal looks like on your machine. Before monitoring begins, the system records thousands of 5-minute behavioral windows during typical use, capturing your natural patterns of system queries, network activity, task scheduling, and more. This personal baseline is what every future process is measured against. The more varied your baseline, the more precisely Guardian understands your normal.

Layer 1a (ETW)

Fast-path: per-process event windows

ETW events from each process are buffered into a short window and scored by the Layer 1a ONNX model. Reconstruction error above the active threshold (shipped preset or a bounded local threshold from a trusted baseline) raises a fast-path flag while activity is still in progress.

Layer 1b (behavioral)

Behavioral monitoring: while it runs

Guardian continuously watches 12 live signals from the kernel: firewall rule changes, scheduled task creation, background service queries, policy modifications, and more. These are measured in rolling 5-minute windows and compared against your personal baseline. A program that suddenly starts behaving at rates your machine has never produced before is a second red flag, regardless of what the program claims to be.

Sorting

How Detections Are Sorted

Not every anomaly is a crisis. Detections escalate through four levels: Monitoring (quiet watch), Under Review (needs attention), Investigation Required (deviation evidence to review), and Action Required (Tier 4): corroborated anomalies with a clear Respond path—guidance and Defender handoff, not automated enforcement.

Details for the nerds →

Free & open source

Core detection runs locally under AGPL-3.0. A free account is required to download the Windows installer from this site.

Download for Windows

Experimental software — not a replacement for Microsoft Defender. Terms · Developers

Questions

Does it work alongside Windows Defender?

Yes. AI Malware Guardian is complementary to Windows Defender, not a replacement. Defender handles known malware via signatures; Guardian handles unknown threats via anomaly detection. Running both together gives you the best coverage.

Will it slow down my PC?

No. The background monitor runs at low priority and uses under 1% CPU on average. The ML models run on compact feature vectors; inference takes microseconds, not milliseconds. There is no performance impact during normal use.

What does "no files leave my machine" mean?

All detection and scoring happens locally on your machine. Expected outbound calls are limited to optional signed app updates and your website account when you download the installer—not behavioral streams, file contents, or process telemetry for cloud scoring.

What Windows versions are supported?

Windows 11 (x64) is currently the supported platform. Windows 10 is not currently supported for production use.

Is the app really free?

Yes. The source is AGPL-3.0. A free account is required to download the Windows installer from this site so we can ship updates responsibly and keep an accurate count of active users.

Can I use it on multiple machines?

Yes. Install Guardian on each Windows PC you want to protect. A free account is required to download the installer from this site; you do not need a separate license per machine in the open-source edition.

What happened to quarantine?

The open-source edition uses Respond guidance only. You remediate with Windows Security or your own tools—Guardian does not move or delete files automatically.

SmartScreen or Defender blocked the installer. Is it safe?

Guardian is not yet signed with a publisher certificate, so Windows may warn on download or flag the monitor as unknown software. Download only from this website (Download or your account page). On SmartScreen, choose More info then Run anyway. If Defender removes the monitor, add an exclusion for %LOCALAPPDATA%\AI Malware Guardian\ and reinstall. Report false positives on GitHub Issues when the repo is public.