Microsoft says it can now score every AI agent's risk continuously and cut its access when that score rises, from stripping write permissions all the way to shutting the agent down. The controls, announced October 19, 2026 inside Microsoft 365 E7 and Project Perception, treat agents as potential insider threats. For solo operators, the pattern matters more than the license: scope access, gate writes, and keep a kill switch.
What Microsoft announced on October 19
On Monday, October 19, 2026, Microsoft introduced security tools that assess each AI agent's risk on an ongoing basis. Microsoft calls the scoring feature Dynamic Agent Risk. Signals include what data an agent can reach and whether its actions are drifting from the task it was given. When the score climbs, an admin can take away the agent's ability to make changes while leaving read access in place, require a person to approve a high-impact action, or stop the agent entirely. SiliconANGLE and The Next Web both reported those capabilities, and both point to Microsoft's own security blog post as the announcement.
The features ride on Microsoft 365 E7 and Project Perception, the security system Microsoft unveiled in July. Microsoft is also adding what it calls Guardian Agents to Project Perception. Those agents investigate why another agent's risk changed and can recommend or take containment steps. GeekWire reported that Microsoft is also raising the number of agent types it can detect on employee computers from 25 to 60, including Anthropic's Claude Code, OpenAI's Codex and GitHub Copilot.
Timing and price: read the fine print
The coverage splits on availability. GeekWire reported that the new features are in private preview, with broader availability planned for Microsoft Ignite, which opens November 17 in San Francisco. SiliconANGLE reported that Microsoft gave no availability dates or pricing for the new controls, and The Next Web said it did not say when the features would arrive. Treat the November timeline as a single-source claim until Microsoft confirms it.
The price GeekWire cited is for the E7 license itself: $99 per user per month as of October 19, 2026. E7 launched in May and bundles Microsoft 365 Copilot, the E5 suite, Entra Suite and Agent 365. None of the coverage we reviewed gave a separate price for the new risk controls, so do not budget for them as a line item yet.
What it means for solo operators
Most solo operators will never buy E7, and nobody running three agents from a laptop is going to open a ticket about Dynamic Agent Risk. The architecture is still the one to copy by hand. The enterprise pattern has four parts: limit what an agent can read, separate read permissions from write permissions, require a human to approve anything that moves money or sends something under your name, and keep a kill switch that does not require a developer. That last one matters most. If your only way to stop a misbehaving agent is to hunt down the process in a terminal, you do not have a kill switch. You have a prayer.
The stakes are not theoretical. Our earlier piece on Schwab and JPMorgan blocking unauthorized agents from moving money made the same point from the banking side: the institutions that hold money are drawing hard lines around agents. A one-person agency with live access to an ad account or CRM is holding the same kind of keys, usually with fewer guardrails and no compliance department to say no.
Our take
Microsoft is right to frame agents as insiders. An agent with credentials, a task and no supervision is precisely what a security team would flag in a human employee, and plenty of small businesses have handed agents the keys with less scrutiny than they'd give a summer intern. After 25 years in digital marketing, including time at Google and in Microsoft's Live Meeting group, I have watched plenty of products trade convenience for control and then act surprised when the control was missing. The lesson holds: default to read-only, and make writes earn their way through.
I also think "Dynamic Agent Risk" is the right idea with a slightly ridiculous name. It sounds like a credit score for robots, which is a fine joke until you have to explain it to a client. Keep the controls, but judge any vendor by whether the kill switch works at 2 a.m. on a Sunday, not by the slide deck. Microsoft's scale is impressive. Your agent stack's blast radius is the number that actually matters.
Sources
GeekWire, "Microsoft unveils tools to monitor and corral rogue AI agents," October 19, 2026: https://www.geekwire.com/2026/microsoft-unveils-tools-to-monitor-and-corral-rogue-ai-agents/
SiliconANGLE, "Microsoft targets the 'agentic insider' with new controls for rogue AI agents," October 19, 2026: https://siliconangle.com/2026/10/19/microsoft-targets-the-agentic-insider-with-new-controls-for-rogue-ai-agents/
The Next Web, "Microsoft adds tools to limit and monitor AI agents in Microsoft 365 E7," October 19, 2026: https://thenextweb.com/news/microsoft-365-e7-ai-agents-dynamic-permissions
Microsoft Security blog, "Agents are your newest insiders," October 19, 2026: https://www.microsoft.com/en-us/security/blog/2026/10/19/agents-are-your-newest-insiders-security-must-move-at-their-speed/