@@ -2,52 +2,42 @@ title: Codename MDASH Overview description: Learn how Codename MDASH - Agentic code scanner uses a multi-model agentic AI system to detect code vulnerabilities with depth and accuracy beyond traditional static analysis. ms.topic: overview-ms.date: 05/26/2026+ms.date: 08/26/2026 ms.custom: references_regions ai-usage: ai-assisted --- -# Codename MDASH overview (private preview)+# Codename MDASH overview (preview) -Codename MDASH - Agentic code scanner is a capability in Microsoft Defender that uses a multi-model agentic AI system to detect code vulnerabilities with unprecedented depth and accuracy. It extends the Defender CLI with agentic code scanning that goes beyond traditional static analysis.--## Capabilities--| Capability | Description |-|---|---|-| AI-powered vulnerability detection | Submits repository code to a multi-model agentic pipeline that identifies vulnerabilities with greater depth than traditional static analysis. |-| Granular confidence scoring | Rates each finding on a 10-level confidence hierarchy — from UNLIKELY to PROVEN — so teams can prioritize with precision. |-| AI-generated code fixes | Uses the `defender fix` command to generate and apply code fixes directly from scan results. |-| Centralized results in Defender portal | Publishes findings to Microsoft Security Exposure Management for organization-wide tracking and triage. |-| CI/CD integration | Runs as part of automated pipelines with SARIF output compatible with GitHub Advanced Security, Azure DevOps, and other tooling. |+Codename MDASH is an agentic code scanner within Microsoft Defender that uses a multi-model agentic AI system to assist security and engineering teams detect and fix code vulnerabilities with unprecedented depth and accuracy.  ## How it works  This service uses a multistage agentic pipeline where specialized AI agents collaborate to find, validate, and prove vulnerabilities: -- **Prepare** — Ranks files by risk using call-graph analysis and code complexity metrics, prioritizing functions most likely to contain vulnerabilities.+- **Prepare** — The system ranks files by risk using call-graph analysis and code complexity metrics, prioritizing functions most likely to contain vulnerabilities. - **Scan** — More than 100 specialized AI agents (for example, injection-auditor, memory-safety-auditor, auth-bypass-auditor) analyze the ranked code using multiple LLMs. Each agent targets a specific vulnerability class. - **Validate** — The system uses taint analysis and type resolution through Language Server Protocol (LSP) servers. A multi-model agentic debate refines confidence and eliminates false positives. - **Dedup** — The system consolidates duplicate findings, producing a final set of unique, actionable vulnerabilities. -## Language support--Codename MDASH operates across codebases generically and supports all major programming languages. It has 100+ agents with different roles and expertise, and we tune it to deepen expertise across languages and environments.+## Key features & capabilities -Codename MDASH is continually adding new agents and expertise to improve both detection and prioritization of findings. Currently, we've trained individual specially tuned expert agents with particularly deep expertise in languages such as C, C++, Java, and C#.+| Capability | Description |+|---|---|+|AI-powered vulnerability detection |Submits code repositories to a multi-model agentic pipeline that identifies vulnerabilities with greater depth than traditional, pattern-matching static analysis. |+|Granular confidence scoring |Rates each finding with a confidence score so teams can prioritize with precision. |+|AI-generated code fixes |Uses the `defender fix` command in the Defender CLI to generate and apply code fixes directly from scan results. |+|Centralized results in Defender portal |Publishes findings to Microsoft Security Exposure Management for organization-wide tracking and triage. |+|Deployment and developer integration|Connects directly to GitHub and Azure DevOps using connecters, triggers on-demand scans locally or in the CI/CD via Defender CLI and integrates with AI coding environments using a dedicated agentic SKILL.| -## Key features+## Language support -| Feature | Description |-|---|---|-| Multi-model agentic architecture | Over 100 specialized agents coordinated across multiple models - not a single LLM prompt. |-| 10-level confidence hierarchy | Granular confidence scoring from UNLIKELY to PROVEN so teams can prioritize with precision. |-| Integrated remediation | The `defender fix` command generates AI-powered code fixes from scan results. |-| Portal integration | Results publish to Microsoft Security Exposure Management for tracking and remediation. |+Codename MDASH operates across codebases generically and supports all major programming languages. It has 100+ agents with different roles and expertise, and we tune it to deepen expertise across languages and environments. Codename MDASH is continually adding new agents and expertise to improve both detection and prioritization of findings. Currently, we've trained individual specially tuned expert agents with particularly deep expertise in languages such as C, C++, Java, and C#.  ## Requirements -For prerequisites, see [Set up agentic code security](ai-code-security-onboarding.md). For information about permissions, see [Security posture – AI code scan](/defender-xdr/custom-permissions-details#security-posture--ai-code-scan).+- For prerequisites, see [Set up agentic code security](ai-code-security-onboarding.md). +- For permissions, see [Security posture – AI code scan](/defender-xdr/custom-permissions-details#security-posture--ai-code-scan).  ## Allow list