Darktrace says it is integrating its / SECURE AI product with Microsoft Agent 365, placing Darktrace-generated AI-agent risk signals in the Microsoft 365 Admin Center’s Agent Registry. For security teams already standardizing on Microsoft’s control plane for agents, the practical appeal is straightforward: an agent inventory that can show behavioral-risk context alongside Microsoft’s native identity, data, and threat signals without requiring analysts to begin every investigation in a separate vendor console.
The announcement, published by Darktrace on August 4, describes an integration rather than a new enforcement product. Darktrace says its signals will help customers spot anomalous or potentially compromised agents, investigate risk in context, and correlate activity across environments. Microsoft’s own Agent 365 documentation confirms that the platform is designed as a centralized registry and governance layer for agents, with security controls spanning Entra, Purview, and Defender.
What Darktrace has not disclosed is at least as important for administrators planning around the news. There is no stated general-availability date, no list of supported agent frameworks or Microsoft environments, no explanation of which Darktrace / SECURE AI SKU is required, and no description of whether the integration supplies only informational signals or can trigger Microsoft enforcement actions. The announcement is a visibility story, not evidence that Darktrace can block an agent’s activity through Agent 365.
Microsoft Agent 365 already offers an agent registry, identity controls, policy templates, security posture data, auditability, and Defender-based detection. Microsoft describes the product as a control plane that can inventory agents, show their usage and health, apply governance during onboarding, and surface risks to security teams in the tools they already use.
Darktrace’s stated contribution is behavioral analysis: prompt-related risk identification, anomaly detection across AI agents, cross-environment correlation, and insight into emerging or unknown threats. That division is sensible. Microsoft’s platform is built to establish ownership, manage agent identities, apply access controls, and enforce governance requirements. A behavior-focused supplier can add value by answering a different question: whether an agent that is technically authorized to act is behaving in a way that warrants scrutiny.
That distinction should shape expectations. A registry can tell an administrator that an agent exists, who owns it, what access it has, and whether it meets required policy. It does not automatically establish that an agent’s sequence of actions is normal for its role. An agent with approved access to SharePoint, Exchange, a CRM system, and external tools could still be manipulated through prompt injection, misused through an overly broad delegation path, or quietly move data in a way that looks legitimate when each event is considered alone.
Darktrace argues that its Adaptive AI can supply the behavioral context missing from static policy and inventory views. The claim is plausible as an operational model, but the company has not published the detection logic, data fields, alert fidelity, retention rules, or false-positive rates for this Agent 365 connection. Administrators should therefore treat the launch as an additional signal source to validate, not as a substitute for least privilege, data-loss prevention, logging, or explicit approval controls.
More materially, Microsoft changed the licensing boundary on July 1. Agent-level security capabilities for Copilot Studio and Microsoft Foundry agents that had previously been available through Defender for Cloud Apps or Defender for Cloud now require an Agent 365-eligible license. Microsoft says tenants without such licensing lose access to agent discovery, posture assessment, threat detection, and certain investigation experiences for those services.
That makes Darktrace’s integration potentially useful for organizations that have already accepted Agent 365 as the administrative foundation. But it also means the vendor’s promise of a “single control plane” is not cost-neutral. A company that lacks an eligible Agent 365 license cannot simply add Darktrace signals to preserve the Microsoft agent-security functionality it lost in July.
Microsoft’s transition guidance also shows why a console-level integration needs close operational review. The company moved agent inventory toward a new
For a SOC, those details determine whether an integration reduces console switching or simply creates a third place to review the same incident.
That absence does not disprove the integration; Darktrace’s announcement establishes that it has been announced. It does mean IT teams cannot yet independently verify the mechanics behind the marketing language. The initial questions for a proof of concept should be mundane and precise:
A conventional access-control decision can establish that an agent was permitted to read a document repository or call an approved business application. It cannot always establish whether an employee’s request, a malicious prompt embedded in a document, or a compromised connector caused the agent to assemble and transmit data in an abnormal way. Detection needs to connect identity, sequence, destination, data type, and historical behavior.
That is where behavioral analytics could become useful, particularly in Windows and Microsoft 365-heavy estates where agents are beginning to operate across Teams, SharePoint, OneDrive, Exchange, business applications, browser-based workflows, and Windows endpoints. The more legitimate systems an agent can reach, the less useful a simple allow-or-deny signal becomes after access is granted.
But the word could is essential. Darktrace has not claimed that its signals will automatically stop agent tool calls, revoke an Entra agent identity, quarantine agent output, or enforce Purview labels. Microsoft continues to own the native governance and enforcement layers through Entra, Purview, and Defender. Darktrace appears to be adding context to the decision-making process surrounding those controls.
Security teams considering the integration should ask Darktrace for a live demonstration using a realistic case: an agent accessing legitimate Microsoft 365 data, then performing an unusual tool invocation, external transfer, or access expansion. The critical test is whether an analyst can see the responsible identity, the agent’s authority chain, the relevant data classification, the behavior deviation, and a clear response action in time to matter.
Darktrace has announced a potentially useful addition to Microsoft’s growing AI-administration stack. The immediate consequence is more limited than the launch language suggests: customers may gain another behavioral-risk lens in the Agent Registry, but they will still need Microsoft’s licensing, native policies, and response controls to turn that visibility into prevention.
What Darktrace has not disclosed is at least as important for administrators planning around the news. There is no stated general-availability date, no list of supported agent frameworks or Microsoft environments, no explanation of which Darktrace / SECURE AI SKU is required, and no description of whether the integration supplies only informational signals or can trigger Microsoft enforcement actions. The announcement is a visibility story, not evidence that Darktrace can block an agent’s activity through Agent 365.
The integration fills a real gap, but only at the investigation layer
Microsoft Agent 365 already offers an agent registry, identity controls, policy templates, security posture data, auditability, and Defender-based detection. Microsoft describes the product as a control plane that can inventory agents, show their usage and health, apply governance during onboarding, and surface risks to security teams in the tools they already use.Darktrace’s stated contribution is behavioral analysis: prompt-related risk identification, anomaly detection across AI agents, cross-environment correlation, and insight into emerging or unknown threats. That division is sensible. Microsoft’s platform is built to establish ownership, manage agent identities, apply access controls, and enforce governance requirements. A behavior-focused supplier can add value by answering a different question: whether an agent that is technically authorized to act is behaving in a way that warrants scrutiny.
That distinction should shape expectations. A registry can tell an administrator that an agent exists, who owns it, what access it has, and whether it meets required policy. It does not automatically establish that an agent’s sequence of actions is normal for its role. An agent with approved access to SharePoint, Exchange, a CRM system, and external tools could still be manipulated through prompt injection, misused through an overly broad delegation path, or quietly move data in a way that looks legitimate when each event is considered alone.
Darktrace argues that its Adaptive AI can supply the behavioral context missing from static policy and inventory views. The claim is plausible as an operational model, but the company has not published the detection logic, data fields, alert fidelity, retention rules, or false-positive rates for this Agent 365 connection. Administrators should therefore treat the launch as an additional signal source to validate, not as a substitute for least privilege, data-loss prevention, logging, or explicit approval controls.
Microsoft has already made Agent 365 the gatekeeper for agent security
The timing matters because Microsoft has been consolidating AI-agent security around Agent 365. Microsoft’s documentation says Agent 365 became generally available for commercial customers on May 1, 2026, and it is licensed per user. It is available as part of Microsoft 365 E7 or as a standalone product, with Microsoft listing the standalone license at $15 per user per month.More materially, Microsoft changed the licensing boundary on July 1. Agent-level security capabilities for Copilot Studio and Microsoft Foundry agents that had previously been available through Defender for Cloud Apps or Defender for Cloud now require an Agent 365-eligible license. Microsoft says tenants without such licensing lose access to agent discovery, posture assessment, threat detection, and certain investigation experiences for those services.
That makes Darktrace’s integration potentially useful for organizations that have already accepted Agent 365 as the administrative foundation. But it also means the vendor’s promise of a “single control plane” is not cost-neutral. A company that lacks an eligible Agent 365 license cannot simply add Darktrace signals to preserve the Microsoft agent-security functionality it lost in July.
Microsoft’s transition guidance also shows why a console-level integration needs close operational review. The company moved agent inventory toward a new
AgentsInfo data source, deprecated older experiences, changed the handling of legacy real-time protection rules, and required some customers to rebuild blocking policies in the new Security for AI policy experience. Darktrace’s announcement does not say how its data maps into those revised workflows, whether it appears in Advanced Hunting, or whether Darktrace findings can be routed into existing Microsoft Sentinel, Defender XDR, or Security Copilot response processes.For a SOC, those details determine whether an integration reduces console switching or simply creates a third place to review the same incident.
The missing technical details are the deployment risk
Darktrace calls itself one of the first security companies to contribute third-party risk signals to the Agent Registry. Microsoft’s public Agent 365 materials confirm that the registry can include Microsoft agents, ecosystem-partner agents, synchronized agents, and self-registered agents. However, Microsoft’s currently published Agent 365 documentation does not identify Darktrace by name or describe the company’s connector, signal schema, onboarding requirements, or support matrix.That absence does not disprove the integration; Darktrace’s announcement establishes that it has been announced. It does mean IT teams cannot yet independently verify the mechanics behind the marketing language. The initial questions for a proof of concept should be mundane and precise:
- Does the connector surface signals for Microsoft Copilot Studio and Foundry agents only, or does it also cover agents built with third-party frameworks and agents discovered through registry synchronization?
- Which Darktrace data is sent into Microsoft 365, and does it include prompt metadata, user identifiers, agent identifiers, tool-call details, sensitive-content classifications, or links back to the Darktrace console?
- Is the integration read-only, does it create actionable Defender incidents, or can it initiate containment, access revocation, or policy changes?
- Which licenses are necessary on both sides, and are there tenant, region, cloud, or data-residency restrictions?
- How are alert ownership, deduplication, retention, and audit trails handled when Microsoft and Darktrace identify the same activity?
Behavioral signals are useful where policy has already passed
The strongest part of Darktrace’s argument is that agents can create risk while staying within permissions granted to them. Microsoft makes the same basic case in its Agent 365 security guidance, identifying over-privileged agents, tool misuse, vulnerable or misconfigured agents, prompt injection, data leakage, and agent sprawl as distinct risks.A conventional access-control decision can establish that an agent was permitted to read a document repository or call an approved business application. It cannot always establish whether an employee’s request, a malicious prompt embedded in a document, or a compromised connector caused the agent to assemble and transmit data in an abnormal way. Detection needs to connect identity, sequence, destination, data type, and historical behavior.
That is where behavioral analytics could become useful, particularly in Windows and Microsoft 365-heavy estates where agents are beginning to operate across Teams, SharePoint, OneDrive, Exchange, business applications, browser-based workflows, and Windows endpoints. The more legitimate systems an agent can reach, the less useful a simple allow-or-deny signal becomes after access is granted.
But the word could is essential. Darktrace has not claimed that its signals will automatically stop agent tool calls, revoke an Entra agent identity, quarantine agent output, or enforce Purview labels. Microsoft continues to own the native governance and enforcement layers through Entra, Purview, and Defender. Darktrace appears to be adding context to the decision-making process surrounding those controls.
What Microsoft 365 administrators should do now
Organizations already running Agent 365 should treat the Darktrace announcement as a reason to inventory their agent-security workflows before enabling another data source. Confirm that every production agent has a named owner or sponsor, bounded permissions, a documented data-access purpose, and a tested offboarding path. Review whether existing Defender alerts, Advanced Hunting queries, Sentinel playbooks, and incident-response procedures account for agent identities rather than only user and service-principal identities.Security teams considering the integration should ask Darktrace for a live demonstration using a realistic case: an agent accessing legitimate Microsoft 365 data, then performing an unusual tool invocation, external transfer, or access expansion. The critical test is whether an analyst can see the responsible identity, the agent’s authority chain, the relevant data classification, the behavior deviation, and a clear response action in time to matter.
Darktrace has announced a potentially useful addition to Microsoft’s growing AI-administration stack. The immediate consequence is more limited than the launch language suggests: customers may gain another behavioral-risk lens in the Agent Registry, but they will still need Microsoft’s licensing, native policies, and response controls to turn that visibility into prevention.
References
- Primary source: Darktrace
Published: 2026-08-04T00:00:00+00:00
Extending AI Security Visibility with Darktrace and Microsoft Agent 365
Darktrace integrates with Microsoft Agent 365 to surface behavioral AI agent risk signals directly within the Microsoft 365 Admin Center.www.darktrace.com - Related coverage: learn.microsoft.com
Microsoft Agent 365 overview | Microsoft Learn
Discover how Microsoft Agent 365 empowers organizations with secure, scalable, and compliant AI agents to enhance productivity and drive measurable outcomes.learn.microsoft.com - Related coverage: learn.microsoft.com
Secure AI agents at scale using Microsoft Agent 365 | Microsoft Learn
How to use Microsoft Agent 365 to secure all of the AI agents in your environment.learn.microsoft.com - Related coverage: darktrace.com
AI/LLM-Generated Malware Used to Exploit React2Shell
Darktrace identified an AI/LLM generated malware sample exploiting the React2Shell vulnerability within its Cloudypots environment. The incident shows how LLM‑assisted development enables low‑skill attackers to rapidly create effective exploitation tools.www.darktrace.com - Related coverage: developer.microsoft.com
Windows Agentic | Microsoft Developer
The secure, open platform for AI & agents. Native support for MCP and Agent workspace on Windows.developer.microsoft.com - Related coverage: support.microsoft.com
Loading…
support.microsoft.com - Related coverage: microsoft.github.io
Microsoft Agent 365 Resources
Curated resources to help you manage and govern your enterprise-scale agents with Agent 365 — the control plane for AI agents.microsoft.github.io
- Related coverage: microsoft.com
Microsoft Agent 365: The Control Plane for Agents
Observe, govern, and secure AI agents confidently with Agent 365. Extend Microsoft 365 and Microsoft Security controls to manage agentic AI at scale.www.microsoft.com
- Related coverage: cdn-dynmedia-1.microsoft.com
- Related coverage: adoption.microsoft.com
Get started With Microsoft Agent 365 in Microsoft 365 admin center
PDF documentadoption.microsoft.com
- Related coverage: download.microsoft.com
- Related coverage: download.microsoft.com
PDF_MSFT_Cloud_architecture_information protection for GDPR.vsdx
PDF documentdownload.microsoft.com
- Related coverage: info.microsoft.com
Loading…
info.microsoft.com - Related coverage: techradar.com
Microsoft reveals Agent 365 - the new and (hopefully) easy way to get a handle on all these new AI agents at work | TechRadar
Say hello to Agent 365www.techradar.com - Related coverage: windowscentral.com
Microsoft doubles down on agentic AI — Agent 365 prepares for a future with over 1 billion agents | Windows Central
Microsoft kicked off Ignite with a major push into agentic AI, unveiling Agent 365 as its newest tool for automating workflows.www.windowscentral.com - Related coverage: itpro.com
Microsoft's new Agent 365 platform is a one-stop shop for deploying, securing, and keeping tabs on AI agents | IT Pro
The new platform looks to shore up visibility and security for enterprises using AI agentswww.itpro.com - Related coverage: itpro.com
Microsoft opens up Entra Agent ID preview with new AI features | IT Pro
Microsoft Entra Agent ID aims to help manage influx of AI agents using existing toolswww.itpro.com - Related coverage: microsoft.com
Loading…
www.microsoft.com - Related coverage: techcommunity.microsoft.com
- Related coverage: microsoftpartners.microsoft.com
Loading…
microsoftpartners.microsoft.com - Related coverage: techcommunity.microsoft.com
- Related coverage: epcgroup.net
Loading…
www.epcgroup.net - Related coverage: licensingschool.co.uk
Loading…
www.licensingschool.co.uk - Related coverage: licensingschool.co.uk
Loading…
www.licensingschool.co.uk - Related coverage: aucotec.com
Loading…
www.aucotec.com - Related coverage: windowscentral.com
Microsoft 365 premium E7 tier is "the First Frontier Suite" | Windows Central
Microsoft’s latest subscription tier promises AI power at $99/month, but analysts are skeptical about its benefits compared to existing plans.www.windowscentral.com - Related coverage: techradar.com
Microsoft 365 is hiking prices for businesses - here's how much it will cost you | TechRadar
Business and government customers, brace yourselveswww.techradar.com - Related coverage: pcgamer.com
Loading…
www.pcgamer.com