Microsoft's new AI agent reverse-engineers malware on its own

Reverse engineering an unknown binary file from scratch is arguably the most grueling task in cybersecurity. Analysts spend hours peeling back obfuscated assembly code just to figure out what a mysterious file actually does.

Microsoft wants an AI agent to do that dirty work.

Microsoft Research, Microsoft Defender Research, and Microsoft Discovery & Quantum introduced Project Ire, an autonomous AI prototype designed to fully reverse engineer and classify software without human intervention.

According to Microsoft, Ire became the first reverse engineer at the company—human or machine—to author a conviction case strong enough to trigger an automatic block for an Advanced Persistent Threat sample, which Microsoft Defender subsequently blocked.

Giving an LLM its own Ghidra toolkit

Automating malware analysis has long hit a wall because code behavior is inherently ambiguous. Reverse engineering protections often look identical on both malicious tools and legitimate commercial software.

Project Ire tackles this by combining large language models with a battery of real-world reverse engineering tools. It taps into Project Freta memory analysis sandboxes, decompilers, and frameworks like Ghidra and angr.

When handed a mystery binary, Ire maps out its control flow graph to form a baseline memory model. From there, the LLM iteratively calls analysis tools through APIs, examining functions responsible for things like process hooking or command-and-control requests.

Crucially, every single deduction is logged into an auditable chain of evidence. Ire even features a validator tool that cross-checks claims against verified rules, rejecting speculative conclusions before issuing a final verdict.

On a test set of public Windows drivers from the Living off the Land Drivers database and Windows Update, Ire achieved 98% precision and 83% recall, misidentifying just 2% of benign files. It was also put to work on nearly 4,000 hard-target files that had been queued up for manual expert review.

Why it matters

Microsoft Defender protects over a billion monthly active devices, and alert fatigue among senior malware analysts is a massive operational bottleneck.

We think Microsoft made a very smart design choice here. Instead of forcing an LLM to guess whether a piece of raw machine code is evil, they turned the model into an orchestrator that drives real decompilers and outputs a transparent, checkable evidence trail.

This fundamentally shifts how Security Operations Center teams handle backlogs. Instead of starting from scratch on unknown binaries, human researchers can review an agent-generated dossier in minutes.

That said, an 83% recall rate means one in six malicious samples still slipped past in testing. Until detection rates climb and agents prove they can resist anti-analysis evasion tactics, Ire is best viewed as a tire-less tier-two assistant rather than a full replacement for human reverse engineers.