EDR rule as a compensating control: what it needs

An EDR rule counts as a mitigation only on hosts where a healthy agent runs it with a blocking action; in alert-only mode it is monitoring. For PHP2Shell (CVE-2026-109618, KEV 26 Oct 2026), CrowdStrike documents prevention of post-exploitation behavior, while CISA's SIGMA rule is a log detection.

What this control does toward each outcome

  1. RemediateLeaves the flaw: The vulnerable software stays. Coverage depends on the agent being installed and healthy on every affected host.
  2. Mitigate now, fix laterLowers exposure: Detects or blocks the post-exploitation behavior or the exploit process on the host.Guidance points here
  3. Accept with a dated exceptionIf the fix date slips: accept the residual risk in a signed, dated, reviewed exception.Also allowed

Decision Table row: A control already sits in front of the vulnerable component (WAF rule, segmentation, access restriction). Group: What can run on the host.

Evidence to keep: Agent coverage report with date; Rule ID and mode; Test record; Exception list.

Coverage decides the class

For most register entries the control is a rule. Here it is the rule plus agent coverage. An affected host with no agent, a stale agent or a sensor in alert-only mode sits outside the control, and an assessor will ask for that list. Pull the agent report for the affected hosts only, not the fleet.

State of an affected hostWhat the rule can doHow to file it
Healthy agent, rule action set to block or preventStops the targeted behavior on that hostMitigation for that host
Healthy agent, rule action set to alertRaises an alert, lets the behavior runDetective control: file it under increased monitoring
No agent, stale agent or sensor disabledNothingUncovered: an open item with an owner and a date
Host cannot run an agent (appliance, embedded device)NothingNeeds another control; see OT and medical devices that cannot be patched

Example with invented counts: 40 hosts run an affected PHP build, 31 report a healthy agent, and 24 of those run the rule with a blocking action. The mitigation covers 24 of 40 hosts (60 percent). Seven hosts are detective only and nine are open. The entry states those three numbers and lists the nine. Filled in for those hosts:

  • Control: EDR rule. Finding: CVE-2026-109618 on 40 hosts running PHP 8.2 to 8.5 below the fixed versions 8.2.35, 8.3.36, 8.4.27 and 8.5.12
  • Mode: block on 24 hosts, alert on 7; uncovered: 9 hosts (no agent, stale agent or sensor off), list attached
  • Pilot test: one Linux pilot host; the record shows a prevented, not only detected, entry
  • Alert owner and runbook: named; exceptions with approvers: the 9 uncovered hosts
  • Ends when: the fixed PHP version is installed on all 40 hosts; review date 29 Oct 2026, the KEV due date

Reading the action on a Linux host

Read the product's own definition of block before filing an entry. Vendors split the same behavior rule into two states. Elastic's documentation for Elastic Defend malicious behavior protection defines them: Detect "Detects malicious behavior on the host and generates an alert. Elastic Defend will not block the malicious behavior." Prevent "Detects malicious behavior on the host, forces the process to stop, and generates an alert" (Elastic Docs, configure an integration policy for Elastic Defend, read 19 Nov 2026). Only the second state is a mitigation. Microsoft's documentation for EDR in block mode in Defender for Endpoint lists Windows as the supported operating system, so that setting does not apply to the Linux servers in the case below.

PHP2Shell runs on Linux web servers. CrowdStrike's post of 21 Oct 2026 says its indicators of attack "detect and prevent" the post-exploitation behaviors it lists, and names no policy setting. In Elastic Security the state is in the alert's rule name: Elastic's endpoint protection rules include "Behavior - Detected - Elastic Defend" and "Behavior - Prevented - Elastic Defend" (Elastic Docs, endpoint protection rules). In any product, open the detection raised by a test of the behavior on one pilot Linux host and check that the record says the process was blocked or prevented, not only detected. If it says detected, the host is in the alert-only row of the coverage table.

A "block" that acts after a detection stops or cleans up a process or artifact; it does not necessarily stop the exploit request that started it. Record what the product's action does and which operating systems it covers.

Worked case: PHP2Shell, CVE-2026-109618

CVE-2026-109618 is a memory-corruption flaw in PHP (CVSS 3.1 base score 9.8) that lets an unauthenticated request lead to remote code execution. It affects PHP 8.2 before 8.2.35, 8.3 before 8.3.36, 8.4 before 8.4.27 and 8.5 before 8.5.12; PHP 8.1 and older get no upstream fix.

Sources: Fastly's PHP2Shell report, the CISA alert (last revised 26 Oct 2026) and the CISA KEV entry, read 19 Nov 2026.
Date and timeEvent
21 Oct 2026, 03:59 UTCFirst request from a PHP2Shell-named tool in Fastly's telemetry
21 Oct 2026, 14:00 UTCPHP releases 8.5.12, 8.4.27, 8.3.36 and 8.2.35, 10 hours 1 minute after that first request (release time scheduled for 14:00 UTC, as listed in Fastly's timeline, which cites the PHP Foundation)
23 Oct 2026CISA updates its alert to add a SIGMA detection rule
26 Oct 2026CISA adds the CVE to KEV, due 29 Oct 2026, forensic triage marked Yes

Two different things were published for the host, and they sit in different classes.

Prevention, vendor-stated. On 21 Oct 2026 CrowdStrike wrote that its existing indicators of attack "detect and prevent common post-exploitation activity, including web shell deployment, suspicious child processes spawned by PHP or web server processes, and malicious command execution" (CrowdStrike, 21 Oct 2026). That is a vendor's claim about its own product. The rules are generic, not written for this CVE. They act after a PHP worker has been made to start a shell, so they limit what follows the exploit request and do not stop the request. The post does not say which hosts have the prevent action switched on, so the coverage count above decides whether it is a mitigation for a given host.

Detection, CISA. CISA's alert, updated 23 Oct 2026, says to "Monitor for suspicious activity, including PHP worker processes launching shells or downloading tools" and points to its SIGMA rule: "Detects PHP2Shell CVE-2026-109618 CVE-2026-109619 RCE Exploitation Attempts", rule ID 741682b9-041e-4a31-b314-c58b15a82cf2, status test, file CISA_SIGMA_261429_r2v1_CVE_2026_109618-109619_PHP2Shell_TLP_CLEAR.yaml at the root of the cisagov/SIGMA_Rules repository (the copy read on 19 Nov 2026 is marked modified 5 Nov 2026). It matches HTTP request patterns, public exploit markers, PHP error messages and reported indicators in log sources. It alerts and blocks nothing, so it is a monitoring entry. The log patterns that rule matches, and the host evidence Fastly lists, are printed on increased monitoring.

What it costs and where it fails

The cost is false positives and agent overhead. NIST SP 800-53 SI-3 says the control must "Address the receipt of false positives during malicious code detection and eradication and the resulting potential impact on the availability of the system." A block rule that stops a legitimate PHP worker spawning a process breaks the application, so a rule tuned to avoid that may be tuned out of usefulness. SI-4 d adds that detections must be analyzed: "Analyze detected events and anomalies;". The rule also fails where the exploit completes before the agent sees a behavior, and where an alert reaches no one.

Check the rule and the agent

  1. Count affected hosts, hosts with a healthy agent, and hosts with the blocking action, as in the example above. Give each gap an owner and a date.
  2. On one pilot host, trigger a harmless test of the behavior on a host your team owns and keep the record: the console entry that shows a block or prevent action and not only an alert.
  3. Confirm someone receives the alert and has a runbook with a time to first look.
  4. Re-run the coverage count after every agent upgrade and every new server build.
  5. Set a review date. SP 800-40 Rev. 4 asks that the response "continues to be in place": no one deactivates it.

What to record

  • Control: EDR rule
  • Finding: <CVE> on <host group>, <software and version>
  • Rule ID and source: <rule name or ID>, as documented by <vendor> on <date>
  • Mode: block or alert (tested on <date>)
  • Affected hosts: <n>; healthy agent: <n>; blocking action: <n>; uncovered list attached
  • Pilot test record: <alert or block ID, date>
  • Alert owner and runbook: <name>, <link>
  • Exceptions with approvers: <list>
  • Ends when: fixed version installed, or review date <date>

The paired detective control is increased monitoring, and the other host-level preventive options are in the register. Network-side blocking is covered under IPS signatures.

Sources

Last reviewed