Your board saw php2shell on LinkedIn. Some of what they read was real. Some of it grew in the retelling. This post separates the two with dated, linked sources, then names what to examine before the next PHP flaw arrives.

php2shell timeline: what happened from 17 to 26 October

Two flaws in PHP core, CVE-2026-109618 and CVE-2026-109619, carry a CVSS 4.0 score of 9.2 each, set by the PHP project. PHP shipped fixes on 21 October 2026: versions 8.2.35, 8.3.36, 8.4.27 and 8.5.12. The record around that date:

  • 17 Oct: AWS’s MadPot honeypot logged an exploitation attempt from one IP address, four days before public disclosure. AWS reports no success for it.
  • 20 Oct: The CVE IDs leaked with a wrong score of 10.0 (PHP Foundation).
  • 21 Oct: AWS saw its earliest PoC-based attempts at about 02:00 UTC, roughly 12 hours before the release. That is the earliest AWS saw, not the start of attempts.
  • 23 Oct: A third bug, CVE-2026-115560, was disclosed with no emergency fix.
  • 24 Oct: Researcher Adel Karimi recorded 35 shell executions on three honeypots running PHP 8.4.26 under Apache mod_php, a favorable setup.
  • 25 Oct: Palo Alto Unit 42 retracted its claim that php2shell hit a European government agency.
  • 26 Oct: Rapid7 and The Register report that CISA added both CVEs to its Known Exploited Vulnerabilities catalog.

Fact versus noise: four claims that traveled

  • “CVSS 10.0.” The leaked score was wrong. The PHP project scores each flaw 9.2.
  • “Millions of attacks.” A secondhand report quoted Cloudflare’s CEO on millions of attempts. Attempts are not compromises. CrowdSec saw 1,187 IPs match exploit patterns from 21 to 25 October and knew of no publicly confirmed compromise “as far as we know.” It counts IP addresses, not attackers, and it is not proof of safety.
  • “A government agency was hit.” Unit 42 retracted this on 25 October. It was “unable to confirm” a link to the php2shell CVEs.
  • “Actively exploited.” This one holds. CISA’s listing and Karimi’s honeypot results both support it.

Even sourced numbers need care. Unit 42’s figure of 22.6 million “potentially vulnerable” instances is undated and rests on its own telemetry. Three questions sort the rest. Was the number measured or estimated? Was it an attempt or a compromise? Which PHP versions does it cover?

The clock starts before the ticket

Waiting for the ticket assumes a grace period between disclosure and the first attack. For the flaws attackers pick, that period is gone. VulnCheck found that 28.96% of 884 exploited vulnerabilities in 2025 were exploited on or before the day their CVE was published, and 23.43% of 495 in the first half of 2026. Cloudflare logged the first Log4Shell exploitation attempt after public disclosure nine minutes later, in 2021. The other side of the clock: Verizon’s 2026 DBIR puts the median time to fully remediate flaws on CISA’s KEV list at 43 days.

Our view: AI will keep the grace period from coming back. That is a view, not a measured trend. No field data yet attributes the change to AI.

Examine what is susceptible before the next PHP flaw

A component is vulnerable when a known flaw exists. It is susceptible when it keeps producing flaws, whether or not a CVE exists today. PHP shows the pattern: CVE-2012-1823 (php-cgi remote code execution) in 2012, CVE-2019-11043 (PHP-FPM remote code execution) in 2019, CVE-2024-4577, which bypasses the 2012 fix, in 2024, and php2shell in 2026. We call this approach Susceptible Components. Five checks, in order:

  1. List every internet-facing PHP runtime and its version.
  2. Mark versions that no longer get updates. PHP says 8.1 and older do not, and offers source backports for 7.4, 8.0 and 8.1.
  3. Update to the fixed versions and restart all workers.
  4. Test first. A reported regression in 8.4.27 and 8.5.12 can crash PHP processes (php-src issue 24488).
  5. Name your next three components with recurring critical CVEs.

Susceptibility tells you where to look first. It does not predict what breaks next.

What to tell your board

Three sentences a CISO can say to the board hold up:

  1. “php2shell is real, and PHP shipped fixes on 21 October. Many of the scary numbers count estimates and attempts, not confirmed breaches.”
  2. “We do not wait for the ticket. Probing began before the fix was public, so we examine what is susceptible now.”
  3. “Before we change production, we predict what the fix will touch. We decide from facts.”

The third sentence is where PIPE™ (Productivity Impact Prediction Engine) fits. It predicts the business impact of a change before the change runs. Ask it the question your board will ask: if we update PHP on every host tonight, what happens?


Sources and checks

Claim in the postSource
CVSS 4.0 score 9.2 each, set by the PHP project. The leaked 10.0 was wrongGHSA-w8hw, GHSA-6wmf, PHP Foundation
Fixes shipped 21 Oct. 8.1 and older no longer get updatesphp.net
17 Oct attempt from one IP. Earliest sighting about 02:00 UTC on 21 OctAWS
Third bug, CVE-2026-115560The Register, 23 Oct
35 shell executions on three honeypots, 24 OctThe Register, 27 Oct
Unit 42 retracts its government-agency update, 25 OctUnit 42, Machine News
CISA adds both CVEs to KEV, 26 OctRapid7, The Register, 27 Oct
Cloudflare’s CEO cites millions of attemptsSecurity Affairs
CrowdSec: 1,187 IPs, no publicly confirmed compromiseCrowdSec
22.6 million potentially vulnerable instancesUnit 42
28.96% of 884 and 23.43% of 495 exploited on or before CVE dayVulnCheck 2025, 1H 2026
Log4Shell: first attempt after disclosure came 9 minutes laterCloudflare
Median 43 days to fully remediate KEV flawsVerizon DBIR 2026
PHP flaw history: 2012, 2019, 2024Bitsight, NVD
Regression in 8.4.27 and 8.5.12Tenable, php-src issue 24488
Vulnerable versus susceptibleReclaim blog