Microsoft official logo (public domain, Wikimedia Commons) — CryptoBriefing brand treatment

New Defender and access controls start AI agents at zero authority and adjust what they can do as risk signals change

Microsoft now treats the AI agents running inside corporate networks as potential insider threats.

On October 19, 2026, the company unveiled a set of security features built to keep enterprise AI agents on a short leash. The headliners are Dynamic Permissions, Dynamic Agent Risk, and Dynamic Conditional Access. Together, they hand agents minimal access at the start and adjust it in real time as risk signals shift.

Dynamic Permissions handles the first part: starting agents from nothing and expanding access only as needed. Dynamic Agent Risk scores how risky an agent looks based on various indicators. Dynamic Conditional Access then uses that score to decide what the agent is allowed to do at any given moment.

Enforcement is graduated rather than all-or-nothing. Microsoft says the system can cut an agent’s ability to write data while still letting it read. It can also require a human to approve specific actions before the agent carries them out.

Microsoft Defender is also getting better at spotting what is out there. The tool can now identify up to 60 different types of AI agents, up from 25 previously. That reflects 35 new agent types added in recent updates.

The features target three specific security headaches: over-privileging, prompt injection, and action drift.

Over-privileging is the classic one. An agent gets far more access than its job requires, usually because broad permissions are easier to set up. When that agent misbehaves or gets hijacked, every one of those extra permissions becomes a liability.

Daily. Free. Join 34,000+ readers across crypto, finance, and policy.

Prompt injection describes attempts to slip malicious instructions to an AI model through the content it reads, tricking it into doing something its operator never intended.

Action drift covers agents that gradually wander from their intended behavior. No single step looks alarming, but the cumulative path ends somewhere nobody signed off on.

The announcement builds on Project Perception, Microsoft’s effort focused on managing external AI threats. Under that umbrella, the company has introduced Guardian Agents, which are designed to provide continuous supervision of other agents.

The new capabilities are integrated with Microsoft 365 E7, priced at $99 per user per month. Microsoft has framed the rollout as part of an expansion of its secure AI capabilities throughout 2026.

For IT and security leaders, the practical shift is from static permissions to continuous evaluation. Traditional access control asks a single question once: is this identity allowed in? The dynamic model keeps asking, adjusting as conditions change.

The pricing also tells a story. Bundling these controls into Microsoft 365 E7 at $99 per user per month places agent governance in the premium tier.

For the wider cybersecurity industry, expanding Defender’s recognition from 25 to 60 agent types sets a visible benchmark for discovery.