Safe agentic AI end-to-end | Microsoft Safety Weblog

Subsequent week, RSAC™ Convention celebrates its 35-year anniversary as a discussion board that brings the safety neighborhood collectively to handle new challenges and embrace alternatives in our quest to make the world a safer place for all. As we glance in the direction of that milestone, agentic AI is reshaping industries quickly as clients remodel to turn out to be Frontier Companies—these anchored in intelligence and belief and utilizing brokers to raise human ambition, holistically reimagining their enterprise to attain their highest aspirations. Our current analysis exhibits that 80% of Fortune 500 firms are already utilizing brokers.¹

On the identical time, this innovation is going on in opposition to a sea change in AI-powered assaults the place brokers can turn out to be “double brokers.” And chief data officers (CIOs), chief data safety officers (CISOs), and safety determination makers are grappling with the ensuing safety implications: How do they observe, govern, and safe brokers? How do they safe their foundations on this new period? How can they use agentic AI to guard their group and detect and reply to conventional and rising threats?

The reply begins with belief, and safety has all the time been the basis of belief. On this agentic period, safety have to be woven into, and round, each layer of the AI property. It have to be ambient and autonomous, similar to the AI it protects. That is our imaginative and prescient for safety because the core primitive of the AI stack.

At RSAC 2026, we’re delivering on that imaginative and prescient with new purpose-built capabilities designed to assist organizations safe brokers, safe their foundations, and defend utilizing brokers and consultants. Fueled by greater than 100 trillion each day indicators, Microsoft Safety helps shield 1.6 million clients, one billion identities, and 24 billion Copilot interactions.² Learn on to learn the way we will help you safe agentic AI.

Safe brokers

Earlier this month, we introduced that Agent 365 will probably be usually out there on Might 1. Agent 365—the management airplane for brokers—provides IT, safety, and enterprise groups the visibility and instruments they should observe, safe, and govern brokers at scale utilizing the infrastructure you have already got and belief. It contains new Microsoft Defender, Entra, and Purview capabilities that can assist you safe agent entry, forestall knowledge oversharing, and defend in opposition to rising threats.

Agent 365 is included in Microsoft 365 E7: The Frontier Suite together with Microsoft 365 Copilot, Microsoft Entra Suite, and Microsoft 365 E5, which incorporates lots of the superior Microsoft Safety capabilities under to ship complete safety in your group.

Safe your foundations

Together with securing brokers, we additionally want to consider securing AI comprehensively. To actually safe agentic AI, we should safe foundations—the programs that agentic AI is constructed and runs on and the people who find themselves creating and utilizing AI. At RSAC 2026, we’re introducing new capabilities that can assist you acquire visibility into dangers throughout your enterprise, safe identities with steady adaptive entry, safeguard delicate knowledge throughout AI workflows, and defend in opposition to threats on the velocity and scale of AI.

Achieve visibility into dangers throughout your enterprise

As AI adoption accelerates, so does the necessity for complete and steady visibility into AI dangers throughout your atmosphere—from brokers to AI apps and companies. We’re addressing this problem with new capabilities that provide you with perception into dangers throughout your enterprise so the place AI is exhibiting up, how it’s getting used, and the place your publicity to danger could also be rising. New capabilities embody:

• Safety Dashboard for AI offers CISOs and safety groups with unified visibility into AI-related danger throughout the group. Now usually out there.
• Entra Web Entry Shadow AI Detection makes use of the community layer to establish beforehand unknown AI purposes and floor unmanaged AI utilization which may in any other case go undetected. Usually out there March 31.
• Enhanced Intune app stock offers wealthy visibility into your app property put in on gadgets, together with AI-enabled apps, to assist focused remediation of high-risk software program. Usually out there in Might.

Safe identities with steady, adaptive entry

Id is the inspiration of contemporary safety, probably the most focused layer in any atmosphere, and the primary line of protection. With Microsoft Entra, you’ll be able to safe entry and ship complete id safety utilizing new capabilities that show you how to harden your id infrastructure, enhance tenant governance, modernize authentication, and make clever entry choices.

• Entra Backup and Restoration strengthens resilience with an automatic backup of Entra listing objects to allow fast restoration in case of unintentional knowledge deletion or unauthorized modifications. Now out there in preview.
• Entra Tenant Governance helps organizations uncover unmanaged (shadow) Entra tenants and set up constant tenant insurance policies and governance in multi-tenant environments. Now out there in preview.
• Entra passkey capabilities now embody synced passkeys and passkey profiles to allow most flexibility for end-users, making it simple to maneuver between gadgets, whereas organizations in search of most management nonetheless have the choice of device-bound passkeys. Plus, Entra passkeys at the moment are natively built-in into the Home windows Good day expertise, making phishing-resistant passkey authentication extra seamless on Home windows gadgets. Synced passkeys and passkey profiles are usually out there, passkey integration into Home windows Good day is in preview. 
• Entra exterior Multi-Issue Authentication (MFA) permits organizations to attach exterior MFA suppliers immediately with Microsoft Entra to allow them to leverage pre-existing MFA investments or use extremely specialised MFA strategies. Now usually out there.
• Entra adaptive danger remediation helps customers securely regain entry with out help-desk friction via automated self-remediation throughout authentication strategies, adapting to the place they’re of their trendy authentication journey. Usually out there in April.
• Unified id safety offers end-to-end protection throughout id infrastructure, the id management airplane, and id risk detection and response (ITDR)—constructed for fast response and real-time choices. The brand new id safety dashboard in Microsoft Defender highlights probably the most impactful insights throughout human and non-human identities to assist speed up response, and the brand new id danger rating unifies account-level danger indicators to ship a complete view of person danger to tell real-time entry choices and SecOps investigations. Now out there in preview.

Safeguard delicate knowledge throughout AI workflows

With AI embedded in on a regular basis work, delicate knowledge more and more strikes via prompts, responses, and grounding flows—usually quicker than insurance policies can sustain. Safety groups want visibility into how AI interacts with knowledge in addition to the power to cease knowledge oversharing and knowledge leakage. Microsoft brings knowledge safety immediately into the AI management airplane, giving organizations clear perception into danger, real-time enforcement on the level of use, and the boldness to allow AI responsibly throughout the enterprise. New Microsoft Purview capabilities embody:

• Expanded Purview knowledge loss prevention for Microsoft 365 Copilot helps block delicate data reminiscent of PII, bank card numbers, and customized knowledge sorts in prompts from being processed or used for net grounding. Usually out there March 31.
• Purview embedded in Copilot Management System offers a unified view of AI‑associated knowledge danger immediately within the Microsoft 365 Admin Heart. Usually out there in April.
• Purview customizable knowledge safety experiences allow tailor-made reporting and drilldowns to prioritized knowledge safety dangers. Out there in preview March 31.

Defend in opposition to threats throughout endpoints, cloud, and AI companies

Safety groups want proactive 24/7 risk safety that disrupts threats early and incorporates them mechanically. Microsoft is extending predictive shielding to proactively restrict affect and cut back publicity, increasing our container safety capabilities, and introducing network-layer safety in opposition to malicious AI prompts.

• Entra Web Entry immediate injection safety helps block malicious AI prompts throughout apps and brokers by imposing common network-level insurance policies. Usually out there March 31.
• Enhanced Defender for Cloud container safety contains binary drift and antimalware prevention to shut gaps attackers exploit in containerized environments. Now out there in preview.
• Defender for Cloud posture administration provides broader protection and helps Amazon Net Companies and Google Cloud Platform, delivering safety suggestions and compliance insights for newly found assets. Out there in preview in April.
• Defender predictive shielding dynamically adjusts id and entry insurance policies throughout lively assaults, lowering publicity and limiting affect. Now out there in preview.

Defend with brokers and consultants

To defend within the agentic age, we’d like agentic protection. This implies having an agentic protection platform and safety brokers embedded immediately into the circulation of labor, augmented by deep human experience and complete safety companies once you want them.

Brokers constructed into the circulation of safety work

Safety groups transfer quickest with focused assist the place and when work is going on. As alerts floor and investigations unfold throughout identities, knowledge, endpoints, and cloud workloads, AI-powered help must function alongside defenders. With Safety Copilot now included in Microsoft 365 E5 and E7, we’re empowering defenders with brokers embedded immediately into each day safety and IT operations that assist speed up response and cut back handbook effort to allow them to deal with what issues most.

New brokers out there now embody:

• Safety Analyst Agent in Microsoft Defender helps speed up risk investigations by offering contextual evaluation and guided workflows. Out there in preview March 26.
• Safety Alert Triage Agent in Microsoft Defender has the capabilities of the phishing triage agent after which extends to cloud and id to autonomously analyze, classify, prioritize, and resolve repetitive low-value alerts at scale. Out there in preview in April.
• Conditional Entry Optimization Agent in Microsoft Entra enhancements add context-aware suggestions, deeper evaluation, and phased rollout to strengthen id safety. Agent usually out there, enhancements now out there in preview.
• Knowledge Safety Posture Agent in Microsoft Purview enhancements embody a credential scanning functionality that can be utilized to proactively detect credential publicity in your knowledge. Now out there in preview.
• Knowledge Safety Triage Agent in Microsoft Purview enhancements embody a complicated AI reasoning layer and improved interpretation of customized Delicate Info Sorts (SITs), to enhance agent outputs throughout alert triage. Agent usually out there, enhancements out there in preview March 31.
• Over 15 new partner-built brokers prolong Safety Copilot with further capabilities, all out there within the Safety Retailer.

Scale with an agentic protection platform

To assist defenders and brokers work collectively in a extra coordinated, intelligence-driven method, Microsoft is increasing Sentinel, the agentic protection platform, to unify context, automate end-to-end workflows, and standardize entry, governance, and deployment throughout safety options.

• Sentinel knowledge federation powered by Microsoft Material investigates exterior safety knowledge in place in Databricks, Microsoft Material, and Azure Knowledge Lake Storage whereas preserving governance. Now out there in preview.
• Sentinel playbook generator with pure language orchestration helps speed up investigations and automate advanced workflows. Now out there in preview.
• Sentinel granular delegated administrator privileges and unified role-based entry management allow safe and scaling administration for companions and enterprise clients with cross-tenant collaboration. Now out there in preview.
• Safety Retailer embedded in Purview and Entra makes it simpler to find and deploy brokers immediately inside present safety experiences. Usually out there March 31.
• Sentinel customized graphs powered by Microsoft Material allow views distinctive to your group of relationships throughout your atmosphere. Now out there in preview.
• Sentinel mannequin context protocol (MCP) entity analyzer helps automate quicker with pure language and harnesses the pliability of code to speed up responses. Usually out there in April.

Strengthen with consultants

Even probably the most mature safety organizations face moments that decision for deeper partnership—a classy assault, a posh investigation, a state of affairs the place seasoned experience alongside your crew makes all of the distinction. The Microsoft Defender Specialists Suite brings collectively expert-led companies—technical advisory, managed prolonged detection and response (MXDR), and end-to-end proactive and reactive incident response—that can assist you defend in opposition to superior cyber threats, construct long-term resilience, and modernize safety operations with confidence.

Apply Zero Belief for AI

Zero Belief has all the time been constructed on three rules: confirm explicitly, use least privilege, and assume breach. As AI turns into embedded throughout your whole atmosphere—from the fashions you construct on, to the info they eat, to the brokers that act in your behalf—making use of these rules has by no means been extra crucial. At RSAC 2026, we’re extending our Zero Belief structure, the complete AI lifecycle—from knowledge ingestion and mannequin coaching to deployment agent conduct. And we’re making it actionable with an up to date Zero Belief for AI reference structure, workshop, evaluation instrument, and new patterns and practices articles that can assist you enhance your safety posture.

See you at RSAC

If you happen to’re becoming a member of the worldwide safety neighborhood in San Francisco for RSAC 2026 Convention, we invite you to attach with us. Be a part of us at our Microsoft Pre-Day occasion and cease by our sales space on the RSAC Convention North Expo (N-5744) to discover our newest improvements throughout Microsoft Agent 365, Microsoft Defender, Microsoft Entra, Microsoft Purview, Microsoft Sentinel, and Microsoft Safety Copilot and see firsthand how we will help your group safe brokers, safe your basis, and show you how to defend with brokers and consultants. The way forward for safety is ambient, autonomous, and constructed for the period of AI. Let’s construct it collectively.

To study extra about Microsoft Safety options, go to our web site. Bookmark the Safety weblog to maintain up with our skilled protection on safety issues. Additionally, comply with us on LinkedIn (Microsoft Safety) and X (@MSFTSecurity) for the most recent information and updates on cybersecurity.

¹Primarily based on Microsoft first-party telemetry measuring brokers constructed with Microsoft Copilot Studio or Microsoft Agent Builder that have been in use over the last 28 days of November 2025.

²Microsoft Fiscal 12 months 2026 First Quarter Earnings Convention Name and Microsoft Fiscal 12 months 2026 Second Quarter Earnings Convention Name