Imagine a careful building inspector reading the blueprints before anyone moves in. The inspector is looking for hidden weak spots that could make the building unsafe. MDASH is similar, but for software. It reads source code, the human-written instructions that tell a program what to do, and looks for vulnerabilities, which are weaknesses that attackers could use to cause harm. It uses artificial intelligence (AI), software that can work through text and code in a reasoning-like way, to help people find problems that are easy to miss.

The short version

MDASH is a codename for a Microsoft Security system that looks through software source code and searches for serious security bugs. Microsoft Security is the part of Microsoft focused on protecting people, devices, and software from digital attacks.

Microsoft describes MDASH as a multi-model agentic scanning system. That phrase sounds heavy, so here is the plain version. It is not one tool doing one simple check. It is an organized group of AI helpers, using more than one AI model, where an AI model is the underlying engine trained to work with words, code, or both.

The goal is to find deep problems in code and explain them clearly, so human security teams and software teams can decide what to do next.

📌
One sentence. MDASH is an AI-assisted code reader that searches for hard security bugs and writes reports for people.

About the name

Some project names are like nicknames on a workshop door. They help people refer to the work before it has a public product name.

MDASH is a codename, which means an internal or project nickname. It is styled after the em dash, the long dash character used in writing. You say it out loud as M dash.

You do not need the punctuation detail to understand the tool. The important part is that MDASH is the name Microsoft uses in its public descriptions of this AI bug-hunting system.

It is more like a team than a single robot

Think of a hospital. A patient does not rely on one person to do every job. One person reads scans, another checks blood tests, another plans treatment, and someone coordinates the whole effort.

MDASH works more like that team than like one all-knowing assistant. It is a harness, meaning a system that coordinates other tools and workers. Microsoft says it coordinates more than one hundred specialized AI agents. An agent is a small AI worker that is given a specific job and the tools needed to do that job.

MDASH is also multi-model, meaning it can use several different AI models instead of depending on only one. That matters because different models can be better at different kinds of reading, checking, or explaining.

MDASH is a harness that coordinates many specialized agents across many models. A Team, Not One MDASHa harness Coordinatesthe effort Many agents100 plus Many modelseach best at
MDASH is a harness that coordinates many specialized agents across many models.

What MDASH reads

A recipe is easier to understand before it becomes a finished cake. You can see each step, each ingredient, and where a mistake may have entered.

MDASH works on source code, the human-written recipe for software. It does not focus on a compiled program, which is software that has already been turned into a form a computer can run more directly.

This matters because source code can show the path of a problem. A weakness may begin in one file, pass through another file, and become dangerous only later. Reading the source gives MDASH a better chance to follow that path.

MDASH reads source code because it shows the path a weakness travels. Source vs Compiled Source codehuman recipe Compiledmachine form Shows the path Hides the path MDASH reads MDASH skips
MDASH reads source code because it shows the path a weakness travels.

What kinds of bugs it is meant to find

Some home problems are obvious, like a broken window. Others are hidden, like a leak that starts behind one wall and shows up in another room weeks later.

MDASH is aimed at deep and complex vulnerabilities. Deep means the problem is not visible from one small piece of code. Complex means several steps may have to line up before the weakness becomes dangerous.

Microsoft describes MDASH as useful for bugs that ordinary scanners may miss. A scanner is a tool that checks software for known signs of trouble. Many scanners are very useful, but some serious bugs are not simple pattern matches.

One example area is memory safety. Memory is the working space a computer uses while a program runs. A memory-safety bug is a mistake where software uses that working space in an unsafe way, which can sometimes let attackers crash a program or take control of it.

Microsoft says it has used MDASH on some of its own large and important software, including Windows, the operating system that runs many personal computers.

  • It looks for problems spread across many steps.
  • It is designed for bugs that are hard for simple checks to catch.
  • It focuses on source code, not on changing live systems.
  • It produces explanations that people can review.
MDASH targets deep, complex bugs that simple scanners often miss. Deep Bugs Deepnot one line Complexsteps line up Memory safetyunsafe workspace
MDASH targets deep, complex bugs that simple scanners often miss.

The most important safety rule: it only looks

A good adviser can study a problem and make a recommendation, but the adviser should not secretly start changing the building.

MDASH is read-only, meaning it reads code and writes findings but does not change the code or the systems running it. It can point out a possible problem, explain why it matters, and suggest a fix.

It does not apply fixes by itself. It does not take independent action inside a customer system. People stay responsible for choosing whether a finding is real, how urgent it is, and how to fix it.

✅
Human control. MDASH is designed to give evidence and suggestions, not to make changes on its own.
MDASH reads, reports, and suggests fixes, but never applies them. Read-Only Reads codeinput Writes findingsoutput Suggests fixno apply
MDASH reads, reports, and suggests fixes, but never applies them.

What decisions this creates for teams

A smoke alarm is useful only if people know where to install it, who listens for it, and what happens when it rings. MDASH is similar. The tool matters, but the process around it matters too.

A security leader, meaning someone responsible for reducing digital risk, has to decide which code is important enough to scan first. That may include software used by many customers, software that handles private data, or software that changes often.

A practitioner, meaning someone who does the hands-on security or software work, has to decide how to review each report. Not every possible bug is automatically a real bug. Teams need to check the evidence, reproduce the issue when they can, and agree on the safest fix.

Teams also need to plan timing. A serious vulnerability may need fast repair. A less urgent problem may be fixed during normal development. MDASH can help find and explain risk, but it does not replace judgment, ownership, or careful testing.

Who built it and who can use it

Some tools are public like a flashlight anyone can buy. Others are service tools offered to certain customers through a managed system.

MDASH was built by a Microsoft team focused on automated code security. Automated means a computer does repeated work with less manual effort. Code security means finding and reducing weaknesses in software instructions.

Microsoft says MDASH is offered to eligible Microsoft Defender customers. Microsoft Defender is Microsoft’s family of security products. Eligible means a customer meets Microsoft’s requirements for access.

It is set up and run through the Microsoft Defender portal. A portal is a website where teams manage a service. In this case, security teams use the portal to manage protection and related security work.

📌
Not a consumer app. MDASH is described as a Microsoft security capability for eligible Defender customers, not as a general app for everyone to download.

Key takeaways

  • MDASH is Microsoft’s AI-assisted system for finding security weaknesses in source code.
  • It works like a coordinated team of many AI agents, not like one single chatbot.
  • It uses multiple AI models, which means it can draw on more than one underlying AI engine.
  • It is built for deep, complex vulnerabilities that may be spread across many files and steps.
  • MDASH is read-only: it reports and suggests, but people decide and make changes.
  • Teams still need judgment, review, testing, and a clear plan for fixing what MDASH finds.

Glossary

Artificial intelligence (AI)
Software that can work through words, patterns, or code in a reasoning-like way.
Agent
A small AI worker given a specific job and the tools needed to do it.
AI model
The underlying AI engine trained to work with information such as text or code.
Harness
A coordinating system that organizes other tools or workers.
Source code
The human-written instructions that tell software what to do.
Vulnerability
A weakness in software that an attacker could use to cause harm.
Scanner
A tool that checks software for signs of security trouble.
Read-only
Able to look at information and report on it, but not change it.
Microsoft Defender portal
A Microsoft website where security teams manage Defender security products and related work.