🔍 Read the full analysis: Could AI Help Power A PHP2Shell Attack? A Look At 25 Palo Alto Servers on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Dream says an attacker used PHP2Shell and a Linux kernel flaw to gain root access to 24 Palo Alto Networks firewalls, then reach a 25th server inside a US government-affiliated network. AI tools were found on the attacker’s infrastructure, but Dream says it has no direct evidence they created the exploit or scripts.
Security company Dream reported that an attacker used two software vulnerabilities to gain root access to 24 Palo Alto Networks firewalls and then compromise a 25th server inside a US government-affiliated network. The attack followed the October 21 release of fixes for the PHP flaws known as PHP2Shell; AI tools were found on the attacker’s servers, but their role in building or running the attack has not been established.
According to Dream’s account, the attacker ran an automated batch against 240 firewalls on the evening of October 23 and obtained root access on 24. The targets were in the United States, Japan, Taiwan and South Korea. Dream said 22 compromised firewalls were left with hidden web shells. The attack used PHP2Shell vulnerabilities CVE-2026-109618 and CVE-2026-109619 alongside Fragnesia, a Linux kernel privilege-escalation flaw identified as CVE-2026-46300.
Dream said PHP2Shell provided a way to run code on the firewall’s management web server, while Fragnesia allowed the attacker to raise privileges to root. On October 24, the attacker reportedly used two firewalls associated with the US government-affiliated organization to reach its internal firewall management server, the 25th compromised system. Dream said the attacker then targeted the organization’s Active Directory, which manages user accounts and access. Palo Alto Networks said its Unit 42 researchers independently verified the attacker’s tools and results.
The timing matters because PHP released fixes on October 21, the same day Palo Alto Networks published an advisory. According to the source account, fixed PAN-OS builds for the affected 12.1 and 12.2 branches were released between October 27 and 29. Until those builds were available, Palo Alto’s recommended protection was to keep the management interface off the public internet. Dream said the attack against each firewall took less than 15 minutes, though the source does not provide independent timing data.
Infographic 01 · PHP2Shell on Palo Alto firewalls
Two days from patch to root.
PHP fixed the bug on 21 October. Fixed firewall software for the attacked branches arrived almost a week later. The attack ran in between.
- 13–23 MayFragnesia goes public, kernel fix followsLinux privilege-escalation bug with a public exploit; CVE-2026-46300.
- 8 OctAttacker sets up an AI platformaiscan configured with DeepSeek and Kimi, per Dream’s timeline.
- 21 OctPHP fixes PHP2ShellPHP 8.5.12, 8.4.27, 8.3.36, 8.2.35. Palo Alto advisory the same day; PAN-OS fixes “upcoming”.
- 23 Oct, evening UTCBatch run: 240 targets, 24 rootedFound via device search engine, filtered to US, Japan, Taiwan, South Korea.
- 24 OctInto a government-affiliated networkTunnel, persistence, internal management server rooted (no. 25), Active Directory attacked.
- 26 OctDream publishes · CISA lists PHP2ShellFederal deadline 29 Oct.
- 27–29 OctFixed PAN-OS builds for 12.2 and 12.1CISA lists Fragnesia on 28 Oct; Unit 42 confirms Dream’s results on 29 Oct.
The fix for PHP existed. The fix for the firewall did not, yet. The open management page was the only door that mattered.Thorsten Meyer · my reading
Sources: Dream Research, 26 Oct 2026 (updated 29 Oct); php.net, 21 Oct 2026; Palo Alto Networks advisory (updated 29 Oct 2026); CISA KEV catalog, version 29 Oct 2026; oss-security, 13 May 2026; Unit 42, 29 Oct 2026. Day count: Thorsten Meyer.
Powered by Thorsten Meyer AI
Exposed Management Interfaces Raised Risk
The report illustrates how a known vulnerability can become an urgent operational risk when management interfaces are reachable from the internet and a second flaw can raise an attacker’s privileges. A firewall is meant to control network access; control of its administrative system can instead give an intruder a path deeper into an organization. In this case, Dream says the attacker moved from perimeter devices to an internal server and then targeted Active Directory.
The incident also highlights the gap that can arise between a vendor advisory and patched software being available for every product branch. Administrators may know a vulnerability is serious before they can install a fixed build. Palo Alto’s advice, as described in the source material, was to restrict access to management pages during that period. The findings do not establish how many other organizations remain exposed, but the reported compromise shows why internet-facing administration deserves prompt review.
AI is part of the story, but not a proven explanation for the breach. Dream found Claude Code installed and the models DeepSeek and Kimi configured in the offensive-security platform aiscan. The company said it had no direct evidence that AI wrote the exploit or attack scripts, and that the operation did not appear fully autonomous. The confirmed security lesson is about exposed services, known flaws and access controls; claims about AI’s contribution remain qualified.
Products Worth Considering
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
LG UltraGear 32GX870A-B 32-inch 4K OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From PHP Fixes to Firewall Builds
Dream’s timeline says the attacker configured aiscan with DeepSeek and Kimi on October 8. Fragnesia had been disclosed in May 2026 with a public exploit and a kernel fix, according to the source material. On October 21, PHP issued updates to several supported versions, while Palo Alto Networks published an advisory and said fixed PAN-OS builds were forthcoming.
Before the reported attack, Dream says a search of internet-connected devices identified about 19,000 PAN-OS firewalls with management pages exposed online. A custom scanner obtained version information from about 5,000; filtering by country and software version produced a set of 240 targets. These are figures from Dream’s account, not a confirmed count of vulnerable devices worldwide. The October 23 batch reportedly attacked five targets at a time. Dream published its findings October 26 and updated them October 29; CISA added both PHP2Shell CVEs to its catalog of exploited vulnerabilities on October 26.
““No direct evidence” shows that AI tools wrote the exploit or scripts.”
— Dream
AI’s Role and Full Impact
It remains unclear whether AI tools contributed to exploit development, target selection, scripting or any other part of the operation. Dream’s discovery of Claude Code and configured language models establishes their presence on the attacker’s servers, not their use in this attack. The company’s assessment was that the operation did not appear fully autonomous.
The source account identifies 25 compromised systems and says the attacker targeted Active Directory, but it does not state what information, if any, was accessed or taken, whether the intruder maintained access beyond the reported period, or how the affected organization responded. It also does not establish the total number of vulnerable firewalls still exposed or whether all 24 devices were patched after the attack.
Patch Builds and Incident Response
Organizations using affected PAN-OS versions should check Palo Alto Networks’ advisories and install the relevant fixed builds. The source account says versions for the affected 12.1 and 12.2 branches arrived between October 27 and 29. Administrators should also verify that management interfaces are not publicly accessible and investigate devices for signs of compromise; applying a patch alone does not establish that an already compromised system is clean.
Further updates may clarify the affected organization’s response, whether the Active Directory environment was breached, and what data or systems were accessed. Dream’s October 29 update and any further statements from Palo Alto Networks or relevant authorities may add detail. Until then, the reported totals and the possible role of AI should be treated according to their stated limits: the incident figures are Dream’s account, while AI assistance remains unproven.
Key Questions
What happened in the reported PHP2Shell attack?
Dream says an attacker combined two PHP vulnerabilities with the Fragnesia Linux kernel flaw to gain root access to 24 Palo Alto firewalls, then reached an internal server inside a US government-affiliated network.
Did AI create the exploit?
That has not been established. Dream found AI coding and language-model tools on the attacker’s infrastructure but said it had no direct evidence they wrote the exploit or scripts.
How many systems were reported compromised?
Dream reported 24 firewalls and one internal firewall management server, for a total of 25 systems. Those figures come from Dream’s investigation.
What protection did Palo Alto recommend before fixed builds were available?
The source account says Palo Alto Networks advised customers to keep the firewall management page off the internet while fixes for the affected PAN-OS branches were pending.
Is the full impact known?
No. The report says the attacker targeted Active Directory, but the source material does not confirm what data was accessed, whether it was taken, or the full extent of any continuing access.
Source: ThorstenMeyerAI.com
