Microsoft introduces Guardian Agents to monitor workplace AI agents
By Automate Basics, written with AI from TechCrunch's original
Microsoft has introduced Guardian Agents, a Project Perception feature intended to monitor the AI agents a company uses. It is aimed at organizations that need oversight of agents with access to workplace data and systems.
What changed
Microsoft has introduced Guardian Agents to monitor the AI agents used within a company, according to TechCrunch. The feature belongs to Project Perception, which previously focused on finding and fixing security weaknesses through agents that simulate attacks, detect problems and address them.
The new focus is on agents a company has chosen to use, not just threats coming from outside. Microsoft’s concern, as reported by TechCrunch, is that an agent may have access to sensitive information and the ability to take consequential actions while behaving unpredictably. TechCrunch also reports separate additions to Microsoft Defender that can detect types of agents employees bring to work, including Claude Code and Codex. Guardian Agents is the monitoring feature in Project Perception; the Defender additions are separate.
Who can use Guardian Agents?
Guardian Agents is intended for companies that use AI agents, particularly where those agents can reach important data or tools, according to TechCrunch’s account of Microsoft’s announcement. The report does not specify which customers can access the feature, whether it is already available to them, which plans include it or what it costs. An introduction of the feature is not enough to establish that a particular team can turn it on today.
For workplace teams, the distinction between monitoring and discovery matters. TechCrunch describes Guardian Agents as a way to watch agents a company uses. It describes the new Microsoft Defender capability as a way to find types of agents employees have brought into the workplace. Teams assessing either capability should ask about the product they actually use, rather than assume both functions come together.
How do I try agent oversight at work?
Agent oversight is worth considering first for a task where an AI agent can read sensitive material or act in another work system. Start by identifying which agents the team uses, what information each can access and what actions each is allowed to take. Those questions reflect the risk Microsoft is raising; they are not setup instructions for Guardian Agents.
If Guardian Agents becomes available to your organization, a useful first check would be whether its monitoring helps the responsible team understand what an agent did during a familiar workflow. Check whether the record makes an unexpected action visible and whether someone can review it before relying on the outcome. TechCrunch reports that Microsoft has also added controls to limit agent access and require human approval for risky decisions, but it does not provide instructions for configuring those controls.
Frequently asked questions
What does Guardian Agents do?
Guardian Agents is a Microsoft feature intended to monitor the AI agents a company uses, according to TechCrunch. It is part of Project Perception. The report frames it as a response to the risk that agents with access to workplace data and tools may take unpredictable actions.
Is Guardian Agents included with Microsoft Defender?
TechCrunch places Guardian Agents in Project Perception, not Microsoft Defender. The report separately describes a new Defender capability for detecting types of agents employees bring into the workplace. It does not say whether the products are sold together or which plans include either feature.
Can my company use Guardian Agents now, and what does it cost?
TechCrunch reports that Microsoft introduced Guardian Agents, but gives no customer availability, plan or pricing details. A company considering it would need to confirm access and cost with Microsoft before planning a rollout. The report does not provide a setup procedure.
For you
Learn to do this with AI, free
Short courses for people who aren't engineers: ChatGPT, Claude, prompting and automation. Every lesson is free, with no card.
Start a free courseFor companies
Make an AI or automation tool?
List it in the directory for $9.99, refunded in full if it isn't approved, or claim its page if it is already here.
Related reading
Wabi opens its AI assistant with shared chats and custom appsWabi has opened its personal AI assistant to everyone. According to TechCrunch, it can organize tasks into separate chats, create small apps within them and let people share a chat with others. Wabi is free for now.
OpenAI outlines ChatGPT coding work for less common languagesOpenAI says it is working to improve ChatGPT across less common programming languages. Its announcement mentions new evaluations for software engineering, SQLite and COBOL migration, but does not say whether a new capability is available.
Google adds meeting recaps to Google Chat after Meet callsGoogle is adding meeting recaps to the Chat conversation associated with a Google Meet call when its AI note-taking feature is active. The recap brings decisions, assigned tasks and links to meeting materials into the conversation where a team can follow up.
Tools in this piece
Written with AI from TechCrunch's original and published after automatic checks: every figure here appears in the original, and no sentence is copied from it. The picture is AI-generated. The original is the authority.
Get the next one by email
One email, most weeks: what changed in AI for people who use it at work, and what to do about it.
Source: Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’, TechCrunch, 19 Oct 2026.