On February 10, 2026, Microsoft published the inaugural issue of its Cyber Pulse report, signed by Vasu Jakkal, Corporate Vice President of Microsoft Security. It has one subject: as AI agents pour into enterprises, how should security teams observe, govern, and protect them? The opening number comes from Microsoft first-party telemetry — more than 80% of Fortune 500 companies already have active AI agents built with low-code/no-code tools. The telemetry window covers the last 28 days of November 2025 and counts agents built with Copilot Studio and Agent Builder.
The report’s weight comes from its data scale and its timing. Agents are no longer a roadmap slide; they are production workloads that hold permissions, call external APIs, and generate output autonomously. Microsoft picked the moment enterprise agent counts crossed a threshold to push observability to the top of the security agenda.
How Fast Agents Spread: Even the Most Regulated Industries Are In
The industry breakdown deserves a close look. Software and tech leads at 16%, followed by manufacturing at 13%, financial institutions at 11%, and retail at 9%. Manufacturing and finance — two of the most heavily regulated sectors — sitting near the top says something important: agent adoption is not a startup toy. It has already penetrated the most conservative IT environments on earth.
The survey backbone is the 2026 Data Security Index: a questionnaire of 1,725 data security decision makers (fielded July 16 through August 11, 2025) plus ten in-depth interviews. Survey and telemetry point to the same conclusion — the adoption curve is steep and the protection curve is flat.
29% of Employees Have Used Unsanctioned Agents
The report’s most quotable everyday number: 29% of employees admit they have used unsanctioned AI agents for work tasks. Jakkal’s line will likely be repeated all year: “You can’t protect what you can’t see, and you can’t manage what you don’t understand.”
What makes agents riskier than classic shadow IT is their shape. An agent inherits its creator’s permissions, acts autonomously on instructions, and produces output outside IT’s line of sight. An agent granted mailbox access is an attack surface that can write its own emails and place its own orders. Coverage from outlets like Security Brief highlighted the report’s warning about ungoverned “AI double agents” — agents that are simultaneously a security exposure and a compliance liability.
Microsoft’s Prescription: Apply Zero Trust to Agents
The core recommendation is to treat AI agents the way you treat employees and service accounts: least-privilege access, explicit verification, and the assumption that compromise will eventually happen. Concretely, the report lays out five foundational capabilities for agent governance:
- Registry: a single source of truth — you cannot govern what you have not inventoried
- Access control: least privilege by default
- Visualization: real-time telemetry into what agents are actually doing
- Interoperability: cross-platform governance, not a single-vendor boundary
- Security: built-in threat protection
The interoperability item is the quietly notable one. Microsoft does not draw the governance line around its own product family; it accepts that enterprise agents will come from many frameworks and vendors, and governance has to span them.
Governance and Security Are Different Jobs — and Both Are Company-Wide
The report deliberately separates two terms that get used interchangeably. Governance answers “who owns this agent and what is it allowed to do”; security handles enforcement and threat detection. They complement each other but cannot substitute for each other — policy without detection is theater, detection without policy is noise.
The third message is organizational. AI risk is an enterprise-level, cross-functional risk: legal, HR, compliance, and the board belong at the table, not just the CISO. Jakkal frames security as an accelerant rather than a brake: “Security is a catalyst for innovation.”
What It Means for Developers and Security Teams
Three actionable moves. First, build registry and telemetry interfaces into agents from day one; retrofitting visibility is always more expensive than shipping it. Second, design permissions like you would for a service account: token lifecycles, scoped-down access, instant revocability. Third, put an agent inventory exercise on the annual security plan — consistent with the shift we flagged in our 2026 opening outlook: the enterprise AI battleground is moving from “which model do we pick” to governance and observability. Microsoft just turned that piece of conventional wisdom into a telemetry-backed number. What remains is the doing.
Sources
- 80% of Fortune 500 use active AI Agents — Microsoft Security Blog
- Cyber Pulse: An AI Security Report — Microsoft Security Insider
- Microsoft warns of AI agent risks in Cyber Pulse brief — Security Brief
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.