Microsoft Bolsters AI Security with New Vulnerability Discovery, Prompt Injection Defenses
Microsoft unveils a suite of security enhancements for June 2026, including an AI-powered vulnerability discovery system and new protections against prompt injection attacks targeting local AI agents.
Microsoft is significantly advancing its security posture for the age of AI with a series of new features and updates announced in its June 2026 security blog. The company's vision centers on making security as ambient and autonomous as the AI it aims to protect, with a focus on strengthening identity, multicloud environments, data protection, and developer workflows.
A key highlight is "Codename MDASH," a novel multi-model agentic scanning system designed to autonomously discover, validate, and assist in remediating complex software vulnerabilities. MDASH employs a panel of specialized AI agents that analyze proprietary code and systems, aiming to surface elusive vulnerabilities more quickly and systematically. This system creates a closed loop, connecting vulnerability discovery, validation, and remediation across Microsoft's security ecosystem, with a private preview now available for early adopters.
Microsoft Defender is also expanding its protective capabilities to encompass local AI agents. The security platform now identifies over 25 types of local AI agents and Model Context Protocol (MCP) servers on managed Windows and macOS devices. Crucially, Defender provides runtime protection against prompt injection attacks, a prevalent threat to AI systems. It can detect and block malicious prompts targeting popular coding agents like GitHub Copilot CLI and Claude Code before they can execute harmful actions, with these capabilities currently in preview.
In the realm of identity security, Microsoft Entra Backup and Recovery is now generally available. This managed service offers always-on backups for identity data, designed to be resilient against deletion or modification. Security teams gain enhanced visibility into tenant changes and can restore core directory objects to previous timestamps, further bolstered by Conditional Access policies to prevent permanent data loss.
Microsoft Defender for Cloud is extending its threat protection to open-source relational databases hosted on Amazon Web Services (AWS) Relational Database Service (RDS). This generally available feature includes built-in threat detection for anomalous access and brute-force attempts, alongside automated sensitive data discovery, enabling better prioritization and response to database risks across hybrid cloud environments.
Further enhancing data security, Microsoft Purview customizable reports are now generally available within Data Security Posture Management (DSPM). These reports offer greater flexibility for tailoring views, analyzing trends, and surfacing critical insights for faster decision-making. Additionally, Defender for Cloud is broadening its multi-cloud coverage for AWS and Google Cloud, adding support for numerous resource types and security recommendations to improve attack surface visibility and risk prioritization.
Finally, a new unified identity risk score aggregates signals from across Microsoft Security into a single, explainable measure of identity risk. This score combines behavioral analysis, access patterns, and threat intelligence to provide a comprehensive view, enabling faster prioritization of urgent threats and automated triggering of protective measures like Conditional Access policies.
These updates underscore Microsoft's commitment to integrating security directly into the tools and platforms developers use, aiming to provide consistent visibility and control from development through runtime, thereby securing the entire AI-powered software lifecycle.