The past week brought a familiar pattern: critical vulnerabilities appeared everywhere, but the real story was rarely the CVE itself.
Windows, AI agents, browsers, enterprise appliances and developer infrastructure all showed the same weakness — systems are becoming interconnected faster than security controls can keep up.
Windows: two zero-days, one destination
Windows Has Two Exploited Zero-Days. Both End at SYSTEM
Two Windows vulnerabilities were already being exploited in the wild. Both can turn a local foothold into SYSTEM-level privileges.
Our take: Windows security increasingly depends on what happens after the attacker gets something small. A low-privileged process should not automatically be the beginning of a straight line to the operating system’s highest authority.
AI agents are becoming software with hands
Google ADK Has a CVSS 10 RCE. AI Agents Are the Attack Surface
AI development frameworks are no longer just about generating text. Agents can call tools, access files, execute workflows and interact with external systems.
Our concern is architectural. Giving an AI agent more permissions makes it more useful — and simultaneously turns every vulnerability around that agent into a much bigger problem.
Security teams are still learning how to protect applications. Now they have to protect applications that can make decisions and perform actions on their own.
Legacy software keeps becoming the front door
Oracle Outside In Has Three Fresh RCEs. Legacy File Parsers Are the Door
Three new RCEs hit Oracle Outside In, a technology designed to process and convert files.
This is exactly the kind of component companies forget about. It sits somewhere in the stack, performs a boring technical task and receives files from somewhere else.
Our take: boring software deserves the same security attention as flashy internet-facing applications. Attackers don’t care whether a component is fashionable. They care whether it parses attacker-controlled data.
Telerik showed how old bugs become new weapons
Telerik’s Padding Oracle Just Became an RCE — And the Exploit Is Public
The Telerik vulnerability had already been patched. Then a working public exploit appeared.
That changes the operational equation immediately.
Our view is simple: patch availability is not the finish line. Public exploitation turns an old vulnerability into a fresh incident-response problem for every organization that postponed the update.
The bigger pattern
This week was not really about Windows, Oracle, Telerik or AI individually.
It was about trust boundaries disappearing.
An AI agent can reach infrastructure. A file parser can become an RCE entry point. A local Windows bug can end at SYSTEM. A patched vulnerability can return to the headlines when someone publishes working exploitation code.
Bugstoday’s takeaway: attack surfaces are expanding faster than most organizations are redesigning their security assumptions.
The next serious breach may not require a spectacular zero-day.
It may only require one forgotten component that everyone assumed was harmless.
Today’s Bugs. Tomorrow’s Breaches.
