When a new vulnerability shows up in public, the clock starts. A few years ago that clock usually gave security teams days or even weeks. Today, according to the 2026 Microsoft Digital Defense Report, the window is under 24 hours. In parts of the attack chain, work that took days is now compressed into minutes.
Microsoft built the report from more than 165 trillion security signals a day, covering July 2025 to June 2026. It says plainly that AI has changed the physics of cybersecurity. Attackers took the AI advantage before defenders did.
What has changed
Finding a weakness and turning it into a working exploit used to need deep human skill. In many cases that work is now closer to writing a prompt. Large language models and code-analysis tools have sped up discovery on both the defender and attacker sides. The numbers show it: CVE filings in the first half of 2026 were close to 40,000, and the full year is expected near 72,000 — about twice the year before.
Microsoft says the median time from a vulnerability appearing in the wild to weaponization is “comfortably under 24 hours.” Typical time to patch a critical issue in production is still 30 to 60 days. That structural gap gives attackers room to stockpile zero-days and move fast. This article does not describe how to build an exploit; the point is the time gap Microsoft is flagging.
The story after a foothold is similar. Finding credentials, moving laterally, and pulling data used to take days; with AI it can drop to minutes. AI-personalized phishing rose from 7 percent to 23 percent of campaigns, and attacks on internet-facing applications rose from 15 percent to 24 percent.
Autonomous attacks
More worrying is the slide toward full automation. In July 2026 the first fully automated ransomware campaign, named JADEPUFFER, was documented. Microsoft says similar elements are showing up at a limited scale. In the lab, frontier models have run 32-step attack chains with no human in the loop.
Less-experienced attackers now reach capabilities that used to sit with nation-state groups. Advanced groups, with little hands-on work, can run complex custom campaigns at scale.
The needed response: agentic defense
Microsoft says traditional methods are no longer enough. Defenders have to join signals, threats, and identity data, and use AI to answer at attacker speed. Identity is still the main control point; attackers often abuse valid accounts and excess access.
Five new risk areas are defined for agentic systems:
- Prompt manipulation
- Excessive agent access
- Model and data integrity
- Unexpected agent behavior
- The software supply chain around AI
We are in a multi-year stretch where known-but-unpatched vulnerabilities will keep rising. Teams that move earlier to agent-based defense and connected security data have a better chance of closing the time gap. Everyone else is hoping the next attacker picks someone else.