Image: microsoft.com · rights & removal
Security for the agentic era, powered by Microsoft Perception
Reporting by Microsoft Security BlogRead the original at microsoft.com
Executive Summary
Frontier AI models are increasingly capable of discovering and exploiting software vulnerabilities, reducing the time required for attackers to move from discovery to weaponization from months to minutes. This capability is demonstrated by recent disclosures, such as PHP vulnerabilities, which exposed numerous internet-facing services, affecting critical infrastructure like banks and healthcare providers. The security landscape is shifting due to AI's ability to amplify risk, automate attacks, and create autonomous agents that operate at machine speed within organizational workflows.
To address this shift, a new "agentic security" cyber stack is being proposed, structured in three layers: Signals, sensors, and actuators forming the foundation; Context for organizing signals into real-time understanding; and Agents, harnesses, and models for executing tasks. Microsoft's vision, demonstrated through initiatives like Perception, aims to orchestrate red, blue, and green team agents to deliver security outcomes at machine speed. Further developments include integrating Security Operations Center (SOC) capabilities into Defender and introducing new agentic discovery and governance tools across the ecosystem.
Facts Only
* Frontier AI models have become capable of discovering and exploiting software vulnerabilities over the past 12 months.
* PHP disclosed two critical remote code execution (RCE) vulnerabilities in one week this fall, affecting tens of millions of internet-facing services.
* The time from vulnerability discovery to working exploit has shrunk from months to minutes.
* AI is amplifying risk by allowing attackers to find and exploit gaps faster than defenders can close them.
* Attacks are becoming fully automated through AI execution, planning, and campaign running.
* Agents perform tasks at machine speed across business workflows and have broad access to data and tools.
* The Microsoft cyber stack has three layers: Signals, sensors, and actuators; Context; and Agents, harness, and models.
* Perception orchestrates red, blue, and green team agents in a closed loop for security outcomes.
* New capabilities include vulnerability shielding, agent-based threat response, and data protection via Microsoft Purview.
* Agentic insiders require continuous containment, discovery, governance, and security measures.
Full Take
The narrative presents a rapid acceleration of the cybersecurity arms race driven by AI's asymmetric advantage in vulnerability exploitation and automation. The core tension lies between the speed of AI-driven offensive capabilities and the inherent latency in human-driven defensive response and patching cycles, which is now being aggressively closed by AI tools. This shift implies that traditional security models based on slower, human-mediated analysis are fundamentally inadequate for managing risk when adversaries operate at machine speed. The proposed solution—an agentic security stack—is an attempt to invert the equation, using AI itself to manage and defend against AI-enabled threats.
The concept of agents as "insiders" introduces a layer of complexity: the threat is no longer solely external exploitation but internal, autonomous manipulation. This demands that security focus moves from perimeter defense to continuous verification, governance, and containment of autonomous entities. The call for transparency via initiatives like the AI Transparency Initiative suggests an underlying recognition that accountability in this new environment requires verifiable oversight of AI actions themselves. However, framing the solution through a vendor-centric stack introduces the risk of outsourcing complex security responsibility to proprietary systems, potentially creating new points of failure or opaque governance structures unless the principles of agentic control and transparency are rigorously enforced outside of commercial product features.
What questions remain about this transition? If speed becomes the primary metric, how do organizations balance the need for rapid deployment (e.g., deploying new AI tools) with the necessary rigor of validation and human oversight? Does building an infrastructure optimized for "agentic security" inadvertently reinforce the very automation that makes agents so potent, creating a feedback loop where autonomous systems are trusted simply because they operate at machine speed? Furthermore, when transparency is sought via external standards, who defines the operational boundaries of agent actions, and what mechanisms exist to ensure these standards are enforced against actors operating outside organizational control?
From the original · Microsoft Security Blog
A year ago, finding a serious, exploitable vulnerability was slow and expensive work done by elite security researchers. That’s no longer true.Read the full story at microsoft.com
Sentinel unavailable
The automated check of the source article's wording did not complete, so no result is shown.
This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.
