Research noteResearched 20 Oct 2026 · published 20 Oct 2026, 12:15 IST

Research note: Microsoft adds risk scores and step-by-step brakes for AI agents, treating them like insiders

How we checked Microsoft’s new agent controls against its own post and two independent reports.

Research led by Mayank Sahu

Compiled with AI assistance (Claude by Anthropic)

Read the story →

At a glance

2searches run

3sources read

3sources used

7facts labelled

Strongevidence strength · 5 of 6 facts (83%)

How we rate evidence: Most facts come from official sources or two or more independent outlets. We count the share of known facts labelled Confirmed or Reported: 70% or more is Strong, 40–69% Mixed, below 40% Limited.

From sources to story

How many sources we read, how many we used, and how many are official or archived.

Read3Every source we opened

Used3Relied on for the story

Cited3Linked in the story

Official1From the company or a public body

Archived3Saved on the Internet Archive

How sure we are, fact by fact

3211

Confirmed · 3 — An official source says it: lab docs, pricing page, notice.
Reported · 2 — Two or more independent, named outlets say it.
Unconfirmed · 1 — One report only, shown with its source.
Not known · 1 — Not announced or not answered yet.

Evidence map

Each row is a fact from the story; each column is a source. A filled dot means that source supports the fact.

Fact	Label	1OFF	2OUT	3OUT
Microsoft unveiled controls that treat AI agents as insider risks on 19 October 2026.	Confirmed
Dynamic Agent Risk gives each agent a continuously updated risk score.	Confirmed
Controls can remove write access, require human approval or stop an agent as risk rises.	Reported
Detection grows from 25 to 60 agent types, including Claude Code, Codex and GitHub Copilot.	Reported
Guardian Agents in Project Perception investigate risk changes.	Confirmed
Features are in private preview with broader availability planned around Ignite.	Unconfirmed
Pricing for the new controls.	Not known

Question

What we set out to check

What the new controls do, which products include them, and when customers can use them.

Findings

What we found

Microsoft’s own post confirms the “agentic insider” approach, Dynamic Agent Risk and Guardian Agents.
SiliconANGLE and GeekWire both describe the graduated brakes and the jump from 25 to 60 detectable agent types.
Only GeekWire gives private preview status and the E7 price.

Discrepancies

Numbers that differ

None. Microsoft’s post itself gives no numbers; the figures come from the two outlets.

Method

Labels we chose and why

Confirmed for features named in Microsoft’s post.
Reported for details in both outlets but not in the post we read.
Unconfirmed for availability timing, seen only in GeekWire.

Open questions

What we could not verify

The separate Microsoft Security blog post with technical detail.
Pricing for the new controls.

Can you help confirm one of these? Send us a source. When we find an answer, we update this note and the story, with the date.

Next steps

How we could improve this

Read Microsoft’s Security blog post and Ignite session details, then confirm availability and pricing.

Research timeline

Research20 Oct 2026
Research note published20 Oct 2026, 12:15 IST
Story published20 Oct 2026, 12:25 IST

What we searched

Microsoft agent risk score shut down rogue agents Ignite private preview October 19
Microsoft Project Perception Guardian Agents risk scoring E7 blog

Every source we read

Source	Used?	Why
Official Microsoft Blog Archived copy	USED	Gallot’s post; confirms Dynamic Agent Risk and Guardian Agents.
SiliconANGLE Archived copy	USED	Feature list and quote.
GeekWire Archived copy	USED	Availability and E7 price.

Sources cited in the story

Research notes show our working. They are not peer-reviewed papers. Data: facts as JSON · How we report