Microsoft Brings Perception to Production. Now Comes the Multivendor Agent Reality Test.
CIOs need a coherent defense for an enterprise filling with agents from competing vendors. Microsoft’s Perception launch makes that ambition more credible. Whether it can unite the whole estate is the question that matters.
Take the argument with you.
Read the publication-ready report. Generated from the canonical News article.

What changed
The enterprise is buying agents by the dozen, and even hundreds today. But it needs someone in command when they need oversight or go wrong.
That is the strategic opening behind Microsoft Perception’s general availability at Ignite on November 17. First unveiled in July and in public preview inside Defender since August, the platform now brings coordinated security agents into production. Four blue-team agents covering alert triage, investigation, detection authoring and threat intelligence are generally available, alongside the Recon Agent. This is a substantial release, rather than the first unveiling of the idea.1
For CEOs and CIOs, the appeal reaches beyond a faster AI operations center. An enterprise adopting agents across sales, finance, development and customer service needs a coherent way to manage and if needed, defend the business they operate. Having every supplier bring its own agents, console and interpretation of risk could leave people doing the hardest integration, management, and governance work themselves.
Perception organizes specialized agents into playbooks: red teams look for weaknesses, blue teams investigate threats, and green teams prioritize exposures and coordinate remediation. An orchestration harness assigns work and connects the results, while people set priorities and approve critical actions. The important change is the unit of automation: a coordinated security outcome, rather than another isolated answer from an assistant.2
The multivendor opening—and its current boundaries
Microsoft is previewing Perception integrations that bring in signals from AWS, Okta and Salesforce and connect playbooks to security monitoring, response automation and IT service management tools through APIs and Model Context Protocol, or MCP. The distinction matters: Perception is generally available, but these integrations are in preview. Risk Reduction and vulnerability shielding are also previews; the latter provides temporary protection while teams prepare a permanent fix.3
Specific support varies by task. Microsoft’s product documentation lists AWS and Okta for attack investigation, and AWS, Okta and Salesforce for detection authoring. That is evidence of a practical path beyond Microsoft’s own products, not proof that every agent can take every action in every vendor’s environment.2
There is a second distinction CIOs should preserve. Perception coordinates security work. Agent 365 supplies the broader agent inventory and governance context, with Entra supplying identity and access controls. Agent 365’s connected-platform expansion includes Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks, Snowflake and others. The expanded Entra Agent ID support is in preview, rolling out through December, and capabilities vary by platform.4
Together, those announcements suggest a broader architecture for governing and defending a mixed agent estate. Perception alone is not a universal manager for all the enterprise’s agents.
Why it matters
One place to run defense is a business requirement
The executive problem is straightforward. When a business process crosses suppliers, responsibility for protecting it cannot stop at each supplier’s console. A CEO needs confidence that someone can see the complete problem, determine who is authorized to act and coordinate the response. CIOs need that confidence without having to replace every useful system they already own.
Consider a hypothetical incident involving an AWS workload, an Okta identity and customer information in Salesforce, with a business agent carrying out part of the workflow. Each system might offer useful evidence. The hard work is establishing whether those observations describe the same attack, deciding which access to revoke and containing the threat without unnecessarily shutting down revenue-producing activity. This is a test scenario for the architecture, not a claim that Microsoft has demonstrated that entire chain as a supported production workflow.
The potential prize is substantial: fewer manual handoffs, less repeated investigation and a clearer record of why action was taken. It could also make scarce security expertise more effective. Experienced analysts could spend more time adjudicating ambiguous incidents and improving defenses, instead of repeatedly assembling context from disconnected queues.
“One place” should therefore mean one coherent incident record, consistent authority and coordinated response. It need not mean forcing every specialist to work in one interface. A common picture is valuable when the people using different tools can still reach and execute a shared decision.
The threat environment makes coordination urgent
Datadog’s November cloud-security research found that the median number of high or critical vulnerabilities per cloud workload rose from four in February to 16 in November across its studied applications in AWS, Azure and Google Cloud. That is a fourfold increase in that workload measure, not a claim that all vulnerability disclosures quadrupled.5
This is the operating pressure behind the agentic security story. If exposure expands faster than teams can investigate and repair it, the organization needs more than better summaries. It needs to choose the right intervention, carry it out within an acceptable time and verify that it worked.
The logic complements the hardening changes covered in our Windows Server analysis: reduce preventable exposure in the underlying estate, then improve how quickly the enterprise handles what remains. Security agents cannot compensate indefinitely for excessive privileges, unowned systems or an inability to deploy fixes.
Visibility is only the first test
CIOs should separate four capabilities that vendor presentations can make look like one: seeing an agent or incident; understanding it; authorizing a response; and enforcing that response in the system where the risk exists.
Microsoft has credible building blocks for that progression. Its security architecture combines signals and enforcement points across Defender, Entra, Intune, Purview and Sentinel with context and agent orchestration. Its broader security ecosystem has more than 500 connectors.3 But connector counts do not establish equivalent response capabilities, permissions or recovery options across all connected products.
Similarly, an agent appearing in a registry does not automatically mean every action it takes is visible or every credential it uses can be revoked through that registry. Agent 365 describes activity visibility on supported platforms and supported identity policies, rather than uniform capabilities everywhere.4 Procurement should preserve those qualifications.
The practical question becomes: for this specific business workflow, which control actually stops the unwanted action? If the answer changes at each vendor boundary, someone still needs to own those differences.
Consolidation changes where the risk sits
A shared defense platform could reduce integration work and give the enterprise a more consistent operating model. It also concentrates dependence. A service disruption, incorrect policy or compromised privileged connection could affect more of the response process at once.
That is a design trade-off, not a reason to reject coordination. Independent emergency access, usable evidence outside the primary console and rehearsed manual containment procedures become more valuable as automation gains authority. The enterprise should be able to stop a dangerous agent even when its preferred management platform is unavailable.
The economics also need a wider lens than analyst hours saved. Count consumption, ingestion, retention, licensing overlap, connector maintenance and the work of reviewing uncertain recommendations. Measure cost per correctly resolved incident and time to verified containment. An inexpensive automated action that disrupts an important business service can be extraordinarily costly.
Frontier take
Microsoft is pursuing the right enterprise problem. The agentic business needs defense that can follow work across organizational and vendor boundaries. A collection of increasingly clever agents with separate permissions and incomplete context will not, by itself, provide that defense.
Our judgment is that Perception makes Microsoft a credible candidate to coordinate this work, particularly where its identity, endpoint and security infrastructure is already deeply deployed. Those existing enforcement points matter: the distance between identifying a problem and doing something about it is central to the value proposition.
The multivendor direction is strategically more important than another increment of intelligence inside a single product. Enterprises should be able to choose business agents for their usefulness while retaining a coherent way to govern the resulting exposure. If Microsoft can make that work reliably across competing platforms, it could earn a larger role in enterprise operations without having to supply every agent itself.
It has not yet demonstrated universal delivery of that ambition. The announced integrations provide a promising starting point; their preview status and uneven scope remain material limits. Coordinating Microsoft security agents using third-party signals is also different from orchestrating arbitrary security agents supplied by every other vendor. Buyers should not collapse those two propositions.
Nor does the combination of Perception, Agent 365 and Entra eliminate the need for a deliberate operating model. Someone must decide which actions agents may take, whose approval matters when priorities conflict and how the enterprise recovers when an authorized action has an unintended result. Security, IT operations and business owners need a shared agreement before the technology can execute one.
This is the security dimension of the larger question in our rolling Ignite analysis: who runs the enterprise once software can exercise meaningful authority? A platform that coordinates defense may become one of its most trusted—and most powerful—operating dependencies.
The opportunity deserves serious investment in evaluation. The purchasing standard should be equally serious: show that the platform can protect the business across the vendors it actually uses, with evidence the enterprise can inspect and authority it can withdraw. Microsoft has strengthened its case. CIOs should now make it prove that case in their own estate.
Three moves for CIOs
-
— Make one mixed-vendor incident the acceptance test Choose a consequential workflow spanning at least two suppliers and run a controlled incident exercise. Require a written capability map covering data access, agent identity, investigation, approvals, enforcement and recovery. Record which steps are generally available, which are preview and which still require a person or another tool.
- Decision trigger: Use the results before expanding privileged access, retiring a competing control or committing to a broad consolidation contract.
- Why now: The headline launch and the integration timetable are different. A successful demonstration inside one vendor’s boundary does not establish reliable protection across yours.
-
— Define authority before delegating response Have the CIO, CISO and relevant business owners agree which actions agents may execute automatically and which require approval. Start with bounded, reversible changes. Assign an accountable owner, capture the evidence behind each decision and test how to stop or reverse an incorrect response.
- Decision trigger: Require review whenever a new integration, privileged permission or business-critical workflow enters the automated response path.
- Why now: The same coordination that can speed containment can spread a mistaken decision. Explicit operating authority makes automation usable without turning every incident into an emergency governance debate.
-
— Buy verified outcomes and preserve an independent escape route Compare the pilot with the current response process using verified containment time, incorrect interventions, total operating cost and analyst effort. Export incident evidence and test emergency access and manual containment during a simulated platform outage. Retire overlapping tools only after their replacement has passed both effectiveness and continuity tests.
- Decision trigger: Apply these gates at renewal and before allowing the platform to become the sole operational route for critical containment.
- Why now: Consolidation can remove expensive duplication, but losing an independent way to act can turn a security-platform failure into an enterprise-wide dependency problem.
Sources
- SecurityWeek: Microsoft Takes Perception Agentic Security Platform Out of Preview 1
- Microsoft: What’s new with Microsoft Perception at Ignite 2026 2
- Microsoft Security: Security for the agentic era, powered by Microsoft Perception 3
- Microsoft: What’s new in Microsoft Agent 365 at Ignite 2026 4
- Datadog: State of Cloud Security, November 2026 5
[^perception][^datadog]