Table Of Content
When the Attacker Is an Agent: Why Patching Alone No Longer Contains the Blast Radius
-
November 2, 2026
-
Patch fast, then assume something got through
When the PHP project shipped emergency releases for two critical request-handling flaws (tracked as PHP2Shell) on 21 October, the response from security vendors was immediate. CrowdStrike and SentinelOne both published customer protection guidance within the day. Two days later, Unit 42 updated its analysis with something more unsettling: evidence that at least one intrusion was carried forward by an attacker whose behaviour it associates with an autonomous AI agent, and that attackers were using agentic AI attack platforms for post-exploitation.
Patching remains step one. But the interesting detail is what happens in the gap between disclosure and patch, and what an agent does once it has a foothold.
Why agents change the arithmetic
A human operator with a new foothold has to look around, decide, and type. An agent explores in parallel, retries when something fails, and chains steps without waiting for a shift change. Zscaler and CrowdStrike have both described attacker tooling built from coding agents, reusable skills and MCP servers earlier this month. The practical effect is a shorter path from “code execution” to “data found and moved”.
That compresses the time defenders have to patch, triage and respond, which is why many teams are moving from “patch everything quickly” to “assume the first control may be late, and limit what can happen next.”
What an agent does after the foothold
Post-exploitation looks familiar: enumerate the host, find credentials and configuration files, read source code and documents, and stage data for transfer. What differs is the volume and speed, and the fact that the activity can look like ordinary scripted tooling.
Notably, much of it happens on a device or server, in process and file activity, before any network control sees a byte. Network inspection is essential, but it sees the transfer, not the search that preceded it.
Containing the blast radius on the device
- 1.Know what is normal for each process. A web service process spawning a shell, or an unfamiliar process reading hundreds of files in a code directory, is a signal regardless of the exploit used.
- 2.Classify data where it lives. If the device already knows which files are source code, secrets, customer records or regulated data, access and movement can be judged by content rather than by filename or path.
- 3.Control egress by content and destination. Block or warn when classified data heads to a new destination, an unsanctioned AI service, or an unknown host, even when the process looks legitimate.
- 4.Use device posture as a gate. Out-of-date or unmanaged systems should have narrower access to sensitive data while patches roll out.
- 5.Preserve the timeline. Process, file, user and network events tied together let responders answer what the intruder touched, not just that they were there.
Keep it proportionate
None of this replaces patching, EDR or network controls. It adds a layer that limits the damage when those controls are late or bypassed, and it works best when alerts are tuned to risky combinations, such as sensitive data plus a new destination, rather than every file read.
Where Kitecyber fits
Kitecyber’s endpoint agent (under 2% CPU) combines device posture, process, file, user and network activity with data classification across 80+ categories. That context lets teams enforce policy on sensitive data movement from the device itself and keep an audit trail of what happened. Deployment typically takes around a day with pre-built compliance policies.
Book a Kitecyber walkthrough at kitecyber.com to see how endpoint data controls limit the blast radius while patches land.
Ajay Gulati
Ajay Gulati is a passionate entrepreneur focused on bringing innovative products to market that solve real-world problems with high impact. He is highly skilled in building and leading effective software development teams, driving success through strong leadership and technical expertise. With deep knowledge across multiple domains, including virtualization, networking, storage, cloud environments, and on-premises systems, he excels in product development and troubleshooting. His experience spans global development environments, working across multiple geographies. As the co-founder of Kitecyber, he is dedicated to advancing AI-driven security solutions.
