Microsoft Admits ‘Frontier Models Can’t Be Trusted’, Unveils New Security Tools to Treat AI Agents as Insider Threats
follow us
Add us as a preferred source on Google
- Microsoft introduces Guardian Agents to monitor, restrict and manage risks posed by autonomous AI agents
- New controls dynamically adjust agent permissions, behavior, and access based on changing risk levels.
- Company warns that “initiated agents” require continuous AI-powered monitoring beyond traditional security controls
AI fanatics, your vocabulary is about to be enriched by two phrases: Agentic Insider and Guardian Agent. These are the two phrases Microsoft used when announcing a series of new tools and features to help companies protect their operations with artificial intelligence agents.
A new Microsoft blog, published by Microsoft corporate vice president of identity and network access Nadim Abdo, argues that agents, while not inherently malicious, can make mistakes or be manipulated and compromised.
Because they need to be granted relatively extensive permissions in order to function properly, this creates a type of risk that can be disastrous if not handled properly.
Latest videos ofTechnologyRadar
Agent Monitoring
Abdo also added that AI agents carrying out their master’s orders cannot be trusted with their own safety.
“Satya Nadella, chairman and CEO of Microsoft, recently described the trust architecture this new era requires: We must separate the provision of intelligence from the authority over it,” Abdo writes.
“We should not ask a model to be its own security authority. The controls that determine what it can access, what actions it can take, and whether it should continue operating should be outside the model, where the model cannot bypass or alter them.”
So supervision is the key word here. But AI agents move at machine speed, something no human can match, and therefore humans are not adequately equipped for the job. Instead, separate AI agents need to be introduced to keep their young ones under control. Previously, Microsoft introduced Project Perception and the Microsoft Cyber stack to help organizations defend themselves with agents.
Project Perception is an AI-powered cybersecurity system designed to coordinate specialized AI agents that analyze security data, identify vulnerabilities, and investigate different threats. The system uses three types of agents: red agents that look for weaknesses, blue agents that investigate potential threats, and green agents that implement corrections and fixes.
Now, Microsoft says it is expanding the project to protect “the insider.” They call him… Guardian Agent.
“Guardian Agents help security teams operate continuously, at machine speed and scale, while sensors, context and actuators across the cyber stack help understand changing agent risk and continually adapt protection in response,” Abdo explained. “Project Perception and Microsoft Agent 365 share signals, context, and a unified log that gives security and IT teams a consistent view of agents and risk, with the controls each needs.”
drill sergeant
Abdo says Guardian Agents is based on a security principle in which non-deterministic intelligence must be surrounded by deterministic controls, independent evidence and containment boundaries.
These controls are divided into seven aspects (identity, permissions, data, endpoints, network, cloud and runtime) and have four responsibilities: knowing which agent exists, who owns it and why, and what it is doing; contain what each agent can access and do before acting; to protect the data, tools, runtime, resources and systems used by the agent; and continually adapt protection as behavior, context and risk change.
In other words, AI needs to be the drill sergeant for drones, adapting to changes in real time to make sure they are doing a good job and not going rogue.
Microsoft also said it is expanding discovery to 35 additional local agent types, including Claude Code, OpenAI Codex, and GitHub Copilot. It also announced the expansion of Microsoft Entra Agent ID with identity types for additional agent patterns, including local and previously unknown agents discovered through security sensors.
With “Dynamic Permissions” and “Dynamic Agent Risk,” Microsoft wants to give companies tighter control over what AI agents can access and do, while controlling their behavior and risk levels. Dynamic permissions limit agents to the access they need for a specific task, while dynamic agent risk scans for potential threats.
Dynamic conditional access can then respond to increasing risks by restricting permissions, blocking certain operations, or requiring human approval. In more extreme scenarios, an agent can be detained and isolated.
The best antivirus for all budgets
Follow TechRadar on Google News and add us as a preferred source to receive news, reviews and opinions from our experts in your feeds.
Post Comment