Microsoft argues that artificial intelligence has changed the pace of cyberattacks to the point where defenders working manually cannot keep up. The claim appears in a Microsoft Security Blog post published around its Ignite 2026 event, where the company presented a set of agent-based security products. The assessment is Microsoft's own, and the post offers no independent confirmation of it.
The PHP example
As its main illustration, Microsoft points to a single week this fall in which PHP disclosed two critical remote code execution flaws. Such bugs allow an attacker to seize a server through one request sent across the internet, the company explained. When the flaws were revealed, tens of millions of internet-facing PHP services ran affected versions, and about 29 million were still exposed nearly two weeks on, according to Microsoft.
The company says security researchers had a working, weaponized exploit for PHP2Shell roughly a day after disclosure. Microsoft says that kind of work once took a specialist team months.
Microsoft also reports that more than 96,000 vulnerabilities have been published to the CVE Program in 2026, against roughly 40,000 in 2024. It adds that only 26% of the vulnerabilities in the CISA Known Exploited Vulnerabilities catalog were remediated in 2025.
What Perception does
Perception is now generally available, Microsoft said. It runs red team agents that simulate attacks, blue team agents that detect and investigate them, and green team agents that fix problems, all driven by playbooks. The company says customers define the guardrails and sign off on high-impact actions.
Alert Triage, Attack Investigation, Detection Authoring and Threat Intelligence are the four blue team agents now generally available. On the red side, the Recon Agent is generally available and the Web App Pen Test Agent is in preview. Among green team agents, the Posture Prioritization Agent is generally available, while the Risk Reduction Agent is in public preview. Vulnerability shielding, also in preview, is designed to give temporary protection to exposed assets until a lasting patch can be applied. Core capabilities of the Integrated Security Operations Center in Microsoft Defender are generally available as well.
Treating AI agents as insiders
Microsoft describes AI agents as the new insiders: they work at machine speed around the clock with wide access, and like people they can err, be manipulated or be compromised. Its discovery tooling now recognizes 60 local agent types, an increase of 35, among them Claude Code, OpenAI Codex and GitHub Copilot. Agents found this way are issued a Microsoft Entra Agent ID, and Conditional Access can be used to govern them.
The company also introduced an AI Transparency Initiative and said it is widening its role in NIST's SI Agent Standards Initiative. More than 40 organizations, among them Accenture, Adobe, KPMG, Mastercard and Volvo Cars, are helping shape the approach, according to Microsoft.





