~/f4n6 $ cat ./php2shell.md
// page

php2shell (CVE-2026-109618 / CVE-2026-109619): live status


Last updated: 2026-11-16T22:02:18Z

Detection package

What this is: the downloadable companion to this page — a self-contained defensive bundle for this incident. It contains the full threat-intelligence dossier, detection rules for Suricata, Zeek, YARA, Sigma and ModSecurity, hunting queries, SOC triage and incident-response guidance, a patch-availability matrix by distribution, and PAN-OS exposure notes. It contains no exploit code.

Version: php2shell-detection-package-2026-10-26 · 47 files · 289 KB · refreshed 2026-11-16 22:02 UTC

SHA256: 124977220781739cc0b09c893d8dfef35ed0c6baabfc215d32ccd3a324ec2443

Download: ZIP, 289 KB · GitHub repository & release notes

The package is rebuilt only when its contents change (checked daily), so the version string always tracks the newest content date.

Summary

Two memory-safety vulnerabilities in PHP itself — not in WordPress, Laravel or any application. Crafted requests reach the vulnerable code before application logic runs, which is why updating plugins, themes or frameworks does nothing.

Fixed in PHP 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35, released 2026-10-21 14:00 UTC.

On 2026-10-24 we reproduced pre-authentication remote code execution with error display disabled — previously the domain of private tooling — against stock PHP 8.4.26. It was verified on both x86_64 and arm64; the deciding condition is Apache mod_php running a single prefork worker, not the architecture. Public tooling is now capable; it is still not universal (details below).

There are two distinct failure modes, and the second is under-appreciated: remote code execution, and silent disclosure of process memory — which can return other requests' passwords, session tokens, API keys and certificates inside a response cookie, without executing anything.

This page is maintained as the situation develops. It contains no exploit code.

Current status

Checked: 2026-11-16 22:02 UTC — automated, hourly

Last substantive change: 2026-11-16 22:10 UTC — Fourth Delft gap open past 24 h: window frozen at 2026-11-15 17:08:49 UTC, 28 h 51 min behind at the 22:00 check on 16 Nov (4 h 48 min at 22:00 on 15 Nov, 13 h 51 min at 07:00 on 16 Nov). Total flat at 89,571, sources 802, code 182, funnel probe 68,282 / 687 with later steps unchanged; everything after the last advance is unmeasured. Separately NVD returned CVE-2026-109618 to Analyzed (refs 12 / 19, scores and CWEs unchanged, workflow housekeeping). Dossier 69.1, 70 and hunting document updated; package rebuilt to 296,269 bytes (sha 12497722).

CISA KEV: listed

EPSS: CVE-2026-109618 99.99% · CVE-2026-109619 99.99%

NVD: CVE-2026-109618 refs 12 · CVE-2026-109619 refs 19

CVE-2026-115560 (third, informational): not yet published to NVD

Palo Alto PAN-OS / Panorama: affected — all 20 hotfix builds released, including the final 11.1.7-h12 on 29 Oct; Panorama uses the same maintenance releases

Exploitation is confirmed. CISA added both CVEs to the KEV catalog on 2026-10-26 with a 2026-10-29 due date, and a researcher recorded 35 working shells across three honeypots on 24 October against Apache mod_php on PHP 8.4.26 (single worker), matching the public PoC. Victim position changed again on 29 October: the vendor is named and the campaign is verified. Dream Group's report now names Palo Alto Networks PAN-OS as the firewall hit by the custom PHP2Shell chain: 25 servers rooted across the US, Japan, Taiwan and South Korea, including three in a US government-affiliated environment, from an exposed management interface (appweb, which also fronts the GlobalProtect portal) plus the Fragnesia kernel LPE. Palo Alto Networks' Unit 42 has independently verified the attacker's tooling and results, and Palo Alto's advisory records active exploitation against internet-exposed management interfaces. No victim organisation has named itself, and we have not seen the vendor IOCs. Unit 42's earlier European-government claim was retracted on 25 October and is separate from this verified campaign. Both stages are KEV-listed: php2shell due 29 October, Fragnesia (CVE-2026-46300) due 31 October. KEV says exploited in the wild, not that your estate was breached.

Update history

This log combines the vulnerability timeline with what we have analysed, published and verified. Entries are tagged incident (vendor or vulnerability development), assessment (our analysis) or published (something we have shipped).

2026-11-16 · [telemetry] — Fourth Delft gap: window now frozen past 24 hours. The data window last advanced to 2026-11-15 17:08:49 UTC and had not moved at the 22:00 check on 16 November, 28 h 51 min later (it was 4 h 48 min behind at 22:00 on 15 November and 13 h 51 min at 07:00 on 16 November, when it was written up). Total 89,571, 802 source IPs, 182 code-or-dropper senders, funnel probe 68,282 / 687, code execution 10,633 / 155, overload 5,928 / 154, dropper 4,728 / 51, all unchanged since the last advance. Everything after 15 November 17:08 is unmeasured: the flat total is missing data, not a quiet network. The gap is recorded in dossier section 69.1 and the hunting document; the end time and increment will be added when the window advances. Package refreshed (295,334 -> 296,269 bytes).

2026-11-16 · [ASSESSMENT] — NVD returned the 109618 record to "Analyzed"; refs now 12 / 19, substance unchanged. After the 15 November reference edit put CVE-2026-109618 at "Modified", NVD moved it back to "Analyzed" at 16:54 UTC and re-touched 109619 at 16:17 UTC. Read directly: references 12 (109618, including the Delft lab) and 19 (109619), NIST v3.1 9.8 / 8.1, vendor v4.0 9.2 / 9.2, CWEs 1325, 416, 755, 824 on both, descriptions unchanged. This is workflow housekeeping, not a new assessment, and patch guidance is unaffected.

2026-11-15 · [telemetry] — Third Delft gap closed: 35 h 07 min, +1 request. The data window stalled at 2026-11-14 05:34:35 UTC and resumed at 2026-11-15 16:41:48 UTC. It was 11 h 25 min behind at the 17:00 check on 14 November (below the threshold, logged only) and was written up as an open gap at 22:00 when it passed 12 hours. On resumption the total moved 89,569 → 89,570 (+1) with source IPs flat at 802 and code-or-dropper senders flat at 182; the funnel reads probe 68,281 / 687, code execution 10,633 / 155, overload 5,928 / 154, dropper 4,728 / 51. +1 across 35 hours is far below the normal rate, so the interval is largely or wholly unmeasured: this cannot distinguish "almost no traffic" from "never replayed", exactly as with the 40.4 hour gap on 4–6 November (+4). Do not read the flat total as a quiet network. Package refreshed (294,906 -> 295,334 bytes).

2026-11-15 · [ASSESSMENT] — NVD moved the 109618 record to "Modified" by adding a 12th reference; nothing of substance changed. CVE-2026-109618 went from "Analyzed" to "Modified" at 01:18 UTC and gained one reference, the Delft Threat Intelligence Lab page (tagged Third Party Advisory, already a source this advisory uses). Read directly from NVD: descriptions, NIST v3.1 9.8, vendor v4.0 9.2 and the CWE set (1325, 416, 755, 824) are unchanged, and CVE-2026-109619 remains Analyzed with 19 references. "Modified" is NVD's editing state, not a re-analysis and not a new authority. Patch guidance is unchanged: 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35, plus session.upload_progress.enabled=0 until the next PHP release.

2026-11-14 · [telemetry] — Third Delft gap open: window frozen past 12 h. The lab's data window last advanced to 2026-11-14 05:34:35 UTC. It was 11 h 25 min behind at the 17:00 check and 16 h 25 min behind at 22:00, so the interval is now recorded as a measured gap rather than a quiet stretch. Figures at the last advance: total 89,569, 802 source IPs, 182 code-or-dropper senders, funnel probe 68,280 / 687, code execution 10,633 / 155, overload 5,928 / 154, dropper 4,728 / 51. Everything after 05:34 on 14 November is unmeasured; the flat total is missing data. The end time and increment will be recorded when the window advances. Dossier section 69.1 and the hunting document carry the detail; package refreshed (294,209 -> 294,906 bytes).

2026-11-12 · [ASSESSMENT] — Guard module v3.2.7 released; claims are the author's, not our tests. The third-party mitigation extension HOSTED-POWER/php_uploadprogress_guard published v3.2.7 at 15:20 UTC, adding session_start()/session_reset() isolation for the unserialize() use-after-free, clearing ext/filter's copy of a dropped form, and letting PHP release cached chunks before the upload-progress headroom guard refuses a request. Its README states that it covers all three unreleased PHP issues by default, including CVE-2026-115560, and that its php2shell layers step aside once PHP is patched. We have not built, installed or tested the module at any version and do not endorse it; the official PHP releases remain the only complete php2shell fix. Recorded as a third-party option for hosts that cannot patch, after source review, checksum verification and staging. PATCH-AVAILABILITY section 11 and dossier §71.1 updated; package refreshed (293,450 -> 294,209 bytes).

2026-11-12 · [patch-tracking] — NCSC-NL adds eight vendor-confirmed fixed scopes; inventory 972 / 212 / 213. PR #84 merged at 11:39 UTC, the first movement since 9 November, recording vendor-published fixes for Cisco URWB and Unified Industrial Wireless (26.2.1.9), QNAP QNE (ADRA) (1.0.7.q327), Xerox AltaLink 82xx/81xx/EC81xx (123.xxx.006.29800) and VersaLink C/B415, C/B625, C/B620 (123.xxx.006.30100), Trend Micro Apex One 2019 (Critical Patch 20113) and Apex Central 2019 (Critical Patch 7351, both Japanese Windows editions), and OroCommerce on OroCloud (patched 22 October). The same review keeps Xerox AltaLink C80xx/B80xx/EC80xx, WorkCentre 3655/EC7836/EC7856, generic Apex One/Apex Central, self-hosted OroCommerce, Observium, Fiery DFE and the Xerox Worldwide Ticketing Tool as Vulnerable, because those vendors published mitigations or dates rather than fixed builds. The inventory read 968 / 215 / 205 and now reads 972 products tracked, 212 vulnerable, 213 fixed (status 12 November 08:56 CET). We confirmed the QNAP advisory text ourselves; the OroCommerce page returned 403 to our fetcher, so that row is vendor-cited rather than independently checked here. PATCH-AVAILABILITY section 8c carries the scope table and the mitigation-is-not-a-fix distinction; dossier §69.6 added; package refreshed (290,824 -> 293,450 bytes).

2026-11-09 · [patch-tracking] — AlmaLinux and Rocky Linux fix scopes land; inventory 968 / 215 / 205. PR #82 (external contributor) merged at 08:58 UTC, 19 minutes after the FortiSOAR merge, adding the AlmaLinux 8/9/10 and Rocky Linux 8/9/10 PHP fix evidence: +20 product scopes and +18 fixed, vulnerable count unchanged. Four Rocky streams are recorded as Investigation rather than fixed. No open PRs or issues remain. One merged FortiSOAR entry carries a calibration point worth reading: Fortinet rates the confirmed impact as denial of service, CVSS 3.1 5.3, which is narrower than the upstream RCE description. PATCH-AVAILABILITY section 5 now lists the fixed builds; dossier §69.5 added; package refreshed (289,497 -> 290,824 bytes).

2026-11-09 · [patch-tracking] — NCSC-NL confirms FortiSOAR fixes; fixed count 185 → 187. PR #83 merged on 9 November at 08:39 UTC, adding vendor-confirmed fixed status for FortiSOAR and FortiSOAR PaaS against both php2shell CVEs. The inventory now reads 948 products tracked, 215 vulnerable, 187 fixed; the product and vulnerable counts did not move. PR #82 (AlmaLinux/Rocky PHP fix status) remains open. PATCH-AVAILABILITY gained a Fortinet subsection and the dossier a new §69.4; detection package refreshed (289,049 → 289,497 bytes).

2026-11-07 · [telemetry] — Second Delft gap closed: 16 h 26 min, +36 requests. The data window stalled at 2026-11-06 19:33:53 UTC and resumed at 2026-11-07 12:00:06 UTC. The total moved 85,982 → 86,018 (+36), source IPs 798 → 799, code-or-dropper senders 181 → 182. The increment is far below the rate seen while the window was advancing normally, so the interval is only partly covered. Both measured gaps are now recorded in the hunting document, and no flat period is described as a quiet network.

2026-11-07 · [ASSESSMENT] — Second Delft telemetry gap recorded, still open. After resuming on 6 November, the lab's data window stalled again at 2026-11-06 19:33:53 UTC. By 07:55 UTC on 7 November the total was unchanged at 85,982 requests (798 source IPs, 181 code-or-dropper senders), about 12.4 hours without records on top of the earlier 40-hour interruption. The hunting document now carries both measured gaps and marks every figure as of the 6 November cutoff. The flat total is missing data and should not be read as a quiet network. Detection package refreshed (288,608 → 288,965 bytes).

2026-11-06 · [INCIDENT] — Cisco marks the URWB fix available and adds Snort coverage. Cisco's advisory moved to v1.11 (6 November 14:21 GMT, still marked Final). The Ultra-Reliable Wireless Backhaul and Unified Industrial Wireless fix 26.2.1.9 is now available, where v1.10 had it planned for November 2026. The advisory also publishes Snort rules 1:67375, 1:67359 and 1:67360. Everything else holds: Expressway remains on 15.5.21 with no date, Duo Access Gateway on 3.2.0, and ISE Portal Builder's fix is deployed with no customer action. The rule that an unlisted Cisco product is not affected still applies. The patch-availability matrix and START-HERE are updated.

2026-11-04 · [ASSESSMENT] — NCSC-NL's inventory now records 18 more fixed scopes, including application images and hosted services. The Dutch inventory added fixed builds for seven container records (EGroupware, Snipe-IT, ProjectSend, Pterodactyl Panel, ownCloud 10 and 11, Invoice Ninja), five hosted services (EGroupware, Snipe-IT, Centreon, MODX, Vtiger) and six RHEL 9 SAP update-service scopes, and extended the Part-DB image record to 2.20.3 and 2.20.4. Totals are now 948 products tracked, 215 vulnerable, 185 fixed. The vulnerable count did not move, so this is added coverage rather than a change in what is affected. Practical value: it tells you whether a downstream image or hosted service has a fixed build, while the patch itself remains the PHP release. Their guidance is unchanged.

2026-11-02 · [ASSESSMENT] — Checked the post-patch PHP follow-ups, and corrected one of our own status lines. The upload-progress guard module shipped v3.2.6 today with a claim worth testing: three PHP issues remain unfixed in any release. Verified against php-src: #24481 (unserialize() re-entrancy use-after-free, no CVE) was closed on 30 October with the fix in source, but no PHP release contains it; #24674 is a further route to the same bug and is open, with fix PR #24751 open against master, unmerged; #24488 (the SimpleXML double-free that 8.4.27 and 8.5.12 introduced) is also fixed only in source; and CVE-2026-115560 remains unfixed. Our clarification document had #24481 as "open" and is corrected. None of this re-opens php2shell: the 21 October releases close both CVEs. The trap is reading "closed" as "fixed for customers" when no release carries the fix. Advice unchanged: patch to 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35, keep session.upload_progress.enabled=0 until the next release, and stage the SimpleXML case. The guard module itself fixed a fiber-isolation bug; we have not tested it and do not endorse it.

2026-11-02 · [ASSESSMENT] — NVD closed its re-analysis without changing the record. Both CVEs moved from "Undergoing Analysis" back to "Analyzed" on 2 November, after the two Siemens advisories were added as references on 30 October (11 references for 109618, 19 for 109619). The descriptions, the NIST v3.1 and vendor v4.0 scores, and the CWE set are unchanged: the records still describe the out-of-memory corruption and stale-global-state conditions, with CWEs 1325, 416, 755 and 824. That matters for one specific reason: the re-analysis did not re-map CVE-2026-109618 to the PHP-CGI PATH_INFO claim, so cgi.fix_pathinfo=0 remains a precaution rather than part of the fix. Patch guidance is unchanged.

2026-11-02 · [detections] — Independent telemetry now shows the delivery side, so the package covers it. TU Delft's Threat Intelligence Lab has been publishing live honeypot and network-telescope data since 20 October: at capture, 83,038 exploitation requests from 787 source IPs, of which 181 IPs delivered code or a dropper. Six new Suricata rules (2202610136–…141, 42 in total) block the reported delivery hosts and paths: 45.74.7[.]227/collect/p2srun.php, 217.60.102[.]5/sh, 176.65.139[.]153/wget.sh, the 92.118.112[.]100 loader paths, the pastebin stage-2 token, and a webhook.site collection UUID. A new hunting note maps the funnel, first sightings, malware hashes and the networks involved, with the caveat that a delivery hit means compromised, not blocked. Source is independent research telemetry, not our verification.

2026-10-25 · [ASSESSMENT] — Ingested CYFIRMA's campaign IoC set (defanged, with provenance): 15 staging addresses (+1 variant), 17 suspected source IPs seen doing web-shell tooling and upload-progress poisoning, a Mirai-style dropper host and URL, four scanner user-agents, a multipart boundary convention, malware hashes, and the JACKALPOC blind-RCE canary. Three new detection rules added — including one that alerts when JACKALPOC appears in a response, which means blind code execution succeeded. The IP lists are ephemeral and partly shared infrastructure: they are for hunting and pivoting, not for blind blocking.

2026-10-26 · [INCIDENT] — Reliability benchmark (n=28) — and a correction to our own claim. Same public exploit build, stock PHP 8.4.26, php.ini-production, display_errors=Off: single-worker 15/15 (100%), multi-worker 2/13 (15%). Our earlier "2 of 3" multi-worker result was small-sample luck. Correct read: multi-worker is no longer an absolute barrier, but exploitation there is unreliable — so worker count is not a security control, yet it still strongly affects an attacker's odds. Patching and session.upload_progress.enabled=0 remain the fixes.

2026-10-26 · [INCIDENT] — CAPABILITY CHANGE: public exploit now clears multi-worker Apache pools. The public php2shell exploit shipped a stability update (worker-count estimation fix). We re-tested against stock PHP 8.4.26 with php.ini-production (display_errors=Off) on a default multi-worker prefork pool (75 processes observed, no pinning): RCE in 2 of 3 attempts, uid=33(www-data). Previously multi-worker pools failed at the leak stage. "Single worker" is no longer a mitigation — default Apache prefork deployments are in scope. Reliability is imperfect (~2/3), the target shape is still mod_php+prefork, and session.upload_progress.enabled=0 still breaks the path. Remediation unchanged: patch and restart every worker.

2026-10-30 · [INCIDENT] — Two vendor-status changes: Cisco's advisory is final, and Siemens' affected list is much wider than the appliance. Cisco v1.10 (30 October, marked Final) corrects the URWB fixed release to 26.2.1.9, still planned for November, and reports the ISE Portal Builder fix deployed with no customer action required. Separately, Siemens' SSA-166980 (V1.2, 29 October) now lists grid and energy-management software as affected: AMIS TS, blueplanet NX3, Energy Engage, and Spectrum Power 3, 4, 7 and TG. For several of those Siemens says no fix is available or planned, so for utility estates this stops being only a firewall-patching question and becomes an exposure-reduction problem. NVD added both Siemens advisories as references to both CVEs on 30 October. We have not tested any Siemens product; this is what the vendor publishes.

2026-10-30 · [ASSESSMENT] — The staging advisory was restructured, and its attribution hardened. CYFIRMA now publishes four dated versions of the research: infrastructure and C2 (22 Oct, 27 addresses), source IPs (23 Oct, 17), scan-node addresses (24 Oct, 21), and a current scan/staging set (29 Oct, 24). The latest version upgrades attribution from "suspected" to high confidence: Korean-speaking operators assessed as North Korean state-nexus in the Lazarus Group ecosystem, with a dual financial and espionage mandate. The claimed target set also grew, from 216 hosts to 581. Our hunting document now follows the advisory's versioning rather than a flattened list, keeps the addresses defanged, and notes that some earlier hash values were hashes of HTTP request bodies rather than files, so they are capture artefacts rather than file indicators. These remain CYFIRMA's assessments from a closed-channel source plus infrastructure correlation, not verified by us, and several addresses are likely compromised third-party hosts.

2026-10-30 · [ASSESSMENT] — The suspected staging infrastructure now has an indicator set. CYFIRMA's advisory has been re-issued with what our earlier coverage lacked: a list of suspected scan and staging addresses (21 new scan nodes, 17 Version 2 addresses and 15 staging addresses still valid), request-body hashes, and operator tradecraft from a closed I2P channel. The claimed target set is 581 hosts, expanded from the original 216 across 10 countries. Two things to hold: the attribution is still suspected, from CYFIRMA's closed-channel source, and some of the listed hashes are of HTTP request bodies, not files, so they only match byte-for-byte captures. Individual addresses overlap with the Potassium loader and the Wiz/Arctic Wolf attack-source lists, which raises confidence in those specific indicators without confirming ownership; some are likely compromised third-party hosts. Added as a hunting document with defanged addresses and a caveat that scanning is not compromise.

2026-10-30 · [INCIDENT] — PAN-OS fixes are complete: all 20 builds released. The final build, 11.1.7-h12, shipped on 29 October, so no PAN-OS or Panorama build remains pending. The advisory also now carries an explicit exploitation line: active exploitation has been observed targeting the management interface when it is exposed to the internet (TLP:AMBER, CVSS-BT 9.2). For defenders this closes the patch question for PAN-OS and leaves two things: verify your management plane is not internet-facing, and remember the kernel stage. Fragnesia (CVE-2026-46300) is KEV-listed with a 31 October deadline.

2026-10-30 · [INCIDENT] — The victim picture is now confirmed and vendor-named. Dream Group updated its report to name Palo Alto Networks PAN-OS as the firewall vendor, and Palo Alto Networks' Unit 42 has independently verified the attacker's tooling and confirmed the reported results. The campaign: 25 servers rooted, three in a US government-affiliated environment, via a custom PHP2Shell exploit against an exposed PAN-OS management interface (appweb, which also fronts GlobalProtect) chained with the Fragnesia kernel LPE. Two caveats still matter: no victim organisation has named itself, and we have not seen the vendor IOCs, which are on Palo Alto's site rather than in Dream's post. Unit 42's earlier European-government claim remains retracted and is a different finding from the verified PAN-OS campaign.

2026-10-29 · [ASSESSMENT] — A competing claim about what CVE-2026-109618 is, and why we are not adopting it. A Full Disclosure post argues the CVE is a PHP-CGI PATH_INFO pointer underflow (the CVE-2019-11043 class) rather than the OOM corruption in the official advisory, and says the public PoCs exploit the wrong bug. We checked: the underlying defect is real, upstream PR php-src#23411 exists and explains that FPM got this guard in 2019 while CGI did not. But the PR is open and unmerged, cgi_main.c shows no such fix, and the 21 Oct release changelog has no CGI entry, so the post's "fixed in 8.4.27" claim does not hold. The CVE mapping remains contested: the advisory, KEV listing, honeypot evidence and exploitation reports all describe the session.upload_progress/OOM chain. So: keep the official actions, and add one cheap precaution. If you serve CGI or FPM with the default cgi.fix_pathinfo=1, set it to 0. That closes a 2019-class path underflow surface for an unpatched, separately-tracked bug. It is not a substitute for patching, and we are not repeating the post's unsourced exposure figures.

2026-10-29 · [INCIDENT] — PAN-OS is nearly done: 19 of 20 hotfixes released, one build left. Palo Alto's advisory removed the ETAs for 10.2.7-h39, 10.2.10-h42, 11.2.4-h23, 12.1.4-h13 and 12.1.10-h3, which brings the released total to 19. Only 11.1.7-h12 remains, with a 30 October ETA. Panorama is covered by the same releases, so the practical position is: every affected branch except one 11.1 build can be patched today, and the interim management-access controls are needed only where that last build has not arrived. Worth repeating that the fix does not remove the kernel requirement from the Dream chain: Fragnesia (CVE-2026-46300) is separately KEV-listed, due 31 October.

2026-10-28 · [ASSESSMENT] — The public ecosystem widened, and our mitigation line needed a caveat. The most visible public research repo on these CVEs (1,889 stars, 642 forks, MIT) documents five different RCE chains, including one it calls zero-dependency pre-auth RCE on Apache mod_php and two targeting PHP-FPM. It also aggregates confirmed wild activity: Potassium, a Mirai-derived DDoS botnet, has been delivered through php2shell since 22 October, with loader URLs, C2 domains, payload hashes and a new multipart boundary (----p2shell). We have tested only one public tool, so those chains are unverified by us, but the claim that FPM is unexploitable is now asserted by two independent projects and is no longer comfortable. The more important correction: our earlier line that disabling session.upload_progress "breaks the path" was too strong. The repo's own analysis agrees it blocks the primary upload vector, but states the Zend-engine OOM corruption has no INI fix. Disabling upload progress reduces attack surface; patching PHP is the complete fix. We added the botnet IOCs, seven Suricata rules and one Sigma rule for the new signatures.

2026-10-28 · [INCIDENT] — The kernel half of the chain is now in KEV too: Fragnesia (CVE-2026-46300), due 31 October. CISA added the Linux kernel privilege-escalation flaw on 28 October with a 31 October due date. This is the bug Dream Group reported chaining after PHP2Shell to get root on 25 appliances, and its KEV description matches the page-cache modification technique they described. The practical consequence: both stages of that real-world chain now carry KEV deadlines, PHP2Shell due 29 October and Fragnesia due 31 October. Patching the PHP runtime alone removes the first stage and leaves the root escalation in place, so kernel patch state matters on appliance estates.

2026-10-28 · [INCIDENT] — The victim picture changed: a vendor reports 25 servers rooted, including a US government-affiliated network. Dream Group published research, based on artefacts found on 25 October, describing 25 rooted servers across the United States, Japan, Taiwan and South Korea, three of them in a US government-affiliated environment. The chain used a custom PHP2Shell exploit against the session.upload_progress mechanism on an undisclosed firewall vendor's appweb management server (PHP 8.0.30), then a Linux kernel LPE, Fragnesia (CVE-2026-46300), for root. Post-exploitation reached Active Directory through a reverse SSH tunnel and a decrypted LDAP service account. Two things to hold in tension: this is the first real-world compromise report for these CVEs, and the vendor, versions, IOCs and victim are all undisclosed, with no independent confirmation. No organisation has named itself. Practical response: treat exposed appliance management interfaces as the demonstrated path in, patch both the PHP runtime and the kernel, and hunt for web shells, reverse tunnels and /etc/pam.d/su tampering. Hunting guide added to the package.

2026-10-28 · [INCIDENT] — Cisco accelerates the URWB fix to November 2026, one hour after our previous note. Cisco's advisory moved from v1.8 to v1.9 (28 October 16:48 GMT) and now plans the Ultra-Reliable Wireless Backhaul fix (27.1.1) for November 2026, four months earlier than the March 2027 date in the previous revision. Everything else holds: ISE Portal Builder offline for remediation, Expressway on 15.5.21 with no date, Duo Access Gateway fixed in 3.2.0, and URWB exploitation still requiring administrator credentials. Worth saying plainly: we published March 2027 about an hour ago, and the vendor then changed it. Planned dates carry a revision number for a reason, and the new advisory watch is what caught this one immediately.

2026-10-28 · [INCIDENT] — Cisco update: URWB fix slips to March 2027, ISE Portal Builder now affected. Cisco's advisory moved to version 1.8 on 27 October. Two changes in scope: the Ultra-Reliable Wireless Backhaul fix (27.1.1) is now planned for March 2027, a month later than the previous estimate, and ISE Portal Builder is now listed as affected with the remedy "offline for remediation" and no fixed release. Expressway remains on 15.5.21 without a date, and Duo Access Gateway is fixed in 3.2.0. URWB exploitation still requires administrator credentials, CVSS 6.6, with a workaround in the bug.

2026-10-28 · [ASSESSMENT] — Correction: Panorama is fixed by the same PAN-OS hotfixes. We had carried "Panorama: affected, no fix date" since 23 October. That was a reading error: Palo Alto publishes one shared hotfix table and no separate Panorama row, and the advisory says the issue applies to "PAN-OS software on PA-Series and VM-Series firewalls and on Panorama (virtual and M-Series)". Panorama runs PAN-OS software and upgrades on the same version numbering (11.2.4-h23, 11.1.7-h12, 10.2.7-h39 and so on), so it is not waiting for a separate vendor date. Both NCSC-NL and Brandefense read it the same way, and NCSC's software list now gives Panorama the same fixed thresholds as PAN-OS. The corrected guidance: upgrade Panorama to the same hotfix as its train. Until then the management-access restrictions still apply. Corrected in the package's patch tables, exposure guide and the status line above.

2026-10-28 · [ASSESSMENT] — Independent CSIRT confirmation of the PAN-OS and Citrix fix tables. NCSC-NL's software status list, updated 28 October, now carries the same PAN-OS thresholds we publish: 14 fixed builds and ETAs for the remaining six (12.1.4-h13, 12.1.10-h3, 11.2.4-h23, 11.1.7-h12, 10.2.7-h39, 10.2.10-h42). It also lists the Citrix NetScaler fixed versions from CTX697263. That is a second CSIRT source agreeing with our tables, which is worth having when someone asks whether the numbers are ours or the vendor's.

2026-10-28 · [INCIDENT] — PAN-OS: three more hotfixes shipped, 14 of 20 builds are now out. Palo Alto's advisory timeline for 28 October removed the ETAs for 11.2.7-h23, 11.2.13-h4 and 11.2.16-h1, so those builds are released. That brings the total to 14, leaving six pending builds, all with 29 or 30 October ETAs: 12.1.4-h13, 12.1.10-h3, 11.1.7-h12, 11.2.4-h23, 10.2.7-h39 and 10.2.10-h42. Panorama still has no fix date. One footnote: the FAQ says 11.2.10-h17 is the build that fixes both the PHP and the Brotli KEV issues, while 11.2.10-h16 fixes only PHP, so estates tracking both should move to h17.

2026-10-28 · [INCIDENT] — Siemens RUGGEDCOM APE1808 joins the product scope, because it runs PAN-OS. NVD's 28 October update added Siemens CPE entries to both CVEs, and Siemens advisory SSA-566512 confirms it: RUGGEDCOM APE1808 devices running PAN-OS are affected in all listed branches, 10.2, 11.1, 11.2, 12.1 and 12.2. There is no separate Siemens firmware fix in V1.0; remediation is the PAN-OS release for the device's branch, which means OT estates hosting PAN-OS on APE1808 fall under the same patch guidance, on slower change windows. Added to the patch-availability table.

2026-10-27 · [ASSESSMENT] — Retested the FPM PoC after its leak fix. Still unverified. The project shipped a commit that targets exactly the failure we saw, replacing a single calibrated leak header with a tiled read, and it now also claims PHP 7.4.33 verified end to end. We rebuilt the lab and retested with the updated client: probe passed, the memory-pressure arm was reached, and the ASLR leak failed the same way across 22 arm requests. Two client versions now fail identically in our Docker lab. The likely reason is that the shipped profiles are calibrated to the author's own PHP build rather than the official container image, so this is a lab difference rather than disproof. The claim remains unverified, and the stated exposure shape is unchanged: Apache mod_php plus prefork is confirmed, FPM is not.

2026-10-27 · [ASSESSMENT] — An FPM-only "full RCE" PoC appeared, and we could not confirm it. A public project claims an nginx + PHP-FPM chain for CVE-2026-109619 under a "default installation" shape. Two things temper that. The working conditions include application behaviours that are not defaults: a session that stores a reference, a reflection of a session value, and an operation that decodes attacker-controlled session data or appends to the session, plus ~5 MB request bodies past nginx's 1-byte-default client_max_body_size of 1m. And the upstream "live FPM" workflow that would exercise the exploit is manual-dispatch only and has never run; regular CI tests only lint and profile schemas. We assembled the lab anyway: the tool's probe passed every precondition on a real PHP 8.4.26 FPM and reached the memory-pressure arm, but three leak attempts and one end-to-end attempt failed at the ASLR leak. That is not disproof (our stack is not the author's), and not confirmation. Our stated exposure shape is unchanged: Apache mod_php + prefork is confirmed; FPM remains unconfirmed. Recorded in dossier §53.

2026-10-27 · [ASSESSMENT] — Attempts versus successes in the honeypot data: roughly 3% of sources converted. Across the two public honeypot reports there are 72 distinct source addresses, 66 in the 26-hour attempt window and 6 in the RCE window, with no overlap. Only two sources produced shells, so source-level conversion is about 3%, and those two were not among the earlier attempt sources. The RCE window shows 52 disclosure cookies leading to 35 shells, while the earlier window had 4,227 access records, 30 OOM triggers and 11 worker crashes with zero shells. A request-level success rate is not derivable from this data: capture selection, incomplete streams and body reuse break the denominator. The practical read is that most scanning is noise; conversion comes from a persistent client retrying the four-stage chain, and confirmation is Apache worker to shell parentage, never a 200, an OOM or a cookie alone.

2026-10-27 · [ASSESSMENT] — The honeypot evidence is public, and it validates one of our detection rules. The primary dataset behind the "35 shells" reporting is now available (0x4D31/honeypot-data/reports/php2shell): 35 execution-stage POSTs correlated with 35 shell starts parented by Apache workers, across three single-worker mod_php targets, from two source address clusters. Requests carried no User-Agent and a ten-character C boundary, which is exactly the constant our Suricata boundary rule (2202610123) keys on — its first in-the-wild validation. The chain is consistent with CVE-2026-109618; the evidence does not establish a separate CVE-2026-109619 path. We added the IOC set, body digests and command patterns (04-hunting/honeypot-rce-2026-10-24.md), plus a no-User-Agent fingerprint rule and an Apache-worker-spawns-shell Sigma rule. The dataset confirms the favourable single-worker case, not multi-worker reliability.

2026-10-27 · [INCIDENT] — CISA KEV: both CVEs listed as actively exploited (added 26 Oct, due 29 Oct). CISA added CVE-2026-109618 and CVE-2026-109619 to the Known Exploited Vulnerabilities catalog on 26 October with a 29 October remediation due date. Separately, a researcher confirmed working in-the-wild RCE: 35 shells across three honeypots on 24 October, against Apache mod_php on PHP 8.4.26 with a single worker, using a four-stage chain that closely matches the public PoC. No named organisation has publicly confirmed a compromise through these CVEs, and the earlier Unit 42 victim claim remains retracted. That distinction matters: KEV tells you exploitation is happening, not that you were breached. Patch against the 29 October date and restart workers.

2026-10-27 · [INCIDENT] — Appliance exposure beyond PAN-OS: Citrix NetScaler fixes are out, Cisco fixes are mostly pending. Citrix advisory CTX697263 (created 22 Oct, updated 27 Oct) confirms NetScaler ADC and NetScaler Gateway are impacted and names fixed versions: 14.1-73.49, 13.1-64.32, plus FIPS lines 14.1-73.49 FIPS and 13.1.37.286. Citrix's interim guidance is to restrict management access, the same control PAN-OS customers need. Separately, Cisco lists Expressway Series (fix 15.5.21, date TBD), Ultra-Reliable Wireless Backhaul (fix planned Feb 2027, and exploitation there needs administrator credentials) and Duo Access Gateway (fixed in 3.2.0). If your estate includes an appliance with an embedded PHP runtime, do not assume distro patching covers it: check the vendor advisory. Full vendor table in PATCH-AVAILABILITY.md section 8c.

2026-10-27 · [INCIDENT] — PAN-OS: two more fixes released ahead of ETA: 12.1.7-h11 and 12.2.4-h1. Both were scheduled for 28 October and are now confirmed by the advisory, bringing the total to 11 released builds. If you run PAN-OS 12.2, 12.2.4-h1 closes this out; on 12.1, 12.1.7-h11 is available now, with 12.1.4-h13 (29 Oct) and 12.1.10-h3 (30 Oct) still pending. Panorama remains affected with no published fix date. Nine builds remain pending, with 11.2.13-h4 the next expected.

2026-10-27 · [INCIDENT] — PAN-OS 10.2 fixes are now officially released: 10.2.13-h27 and 10.2.16-h12. The advisory was updated 27 October and removed the ETAs for both builds. If you run PAN-OS 10.2.13, upgrade to 10.2.13-h27 or later; on 10.2.16, upgrade to 10.2.16-h12 or later. 10.2.18-h13 also covers both branches. The remaining eleven pending builds keep their 27 to 30 October ETAs, and Panorama still has no published fix date. After upgrading, re-check management-plane exposure rather than assuming the patch closes an incident.

2026-10-26 · [INCIDENT] — PAN-OS 10.2 branch: documentation is live before the advisory. The "Addressed Issues" pages for 10.2.13-h27 and 10.2.16-h12 are published and state they include the fix for CVE-2026-109618, while the advisory still showed ETAs of 10/27 and 10/28 at 23:59 UTC. These are the only two of the thirteen pending builds with documentation live (11.1.7-h12 is not yet published). Documentation is not a download: check your support portal for availability. Do not wait for the advisory to flip if your portal offers the build, and do not treat the ETA as evidence that it is unavailable. The monitor now HEAD-checks every pending build's release-notes page so documentation publication surfaces as its own signal.

2026-10-26 · [INCIDENT] — Correction: WildFire WF-500 / WF-500-B are NOT affected by php2shell. We had carried "Panorama and WF-500 have no fix date" since 23 October, based on an advisory revision that briefly listed the appliances as in scope. Palo Alto corrected that the same day: its product table now lists WF-500 / WF-500-B as None affected / All unaffected. Our line was stale, and customers with WF-500s were being told to wait for a fix that is not needed. Corrected across the PAN-OS exposure, patch-availability and start-here documents. Panorama is still affected, still with no published fix date. The monitor now parses product-status rows so a vendor correction of this kind surfaces automatically.

2026-10-26 · [ASSESSMENT] — Detection aligned with what the frontier learned: boundary-first, not payload strings. Fastly's telemetry (~250,000 requests) shows there is no payload string — the exploit is a condition — and the reliable discriminator is the fixed/patterned multipart boundary per exploit kit. We added boundary rules (kit boundary B7kQpoc618; fixed single-character boundaries such as CCCCCCCCCC from the public tool), a ----wp2shell look-alike exclusion, and — importantly — excluded React2Shell traffic from our chunked-multipart rules, because that shape is 98.6% React2Shell (CVE-2025-55182) and the A-padding variant is 99.9%. UA, IP blocklists and body-size thresholds are documented red herrings. Confirmation cannot be made at the edge: a successful attempt looks identical to a failed one — check origin logs (PHP-FPM, Apache AH00051 worker crashes) and connect-back egress.

2026-10-26 · [INCIDENT] — PAN-OS hotfixes have started shipping. Palo Alto's advisory timeline records seven released builds: 10.2.18-h13, 11.1.16-h4, 11.2.10-h16 (25 Oct) and 11.1.4-h38, 11.1.6-h40, 11.1.10-h36, 11.1.13-h16 (26 Oct); the release notes for those builds state they include the fix for CVE-2026-109618. 13 further builds remain on ETA 27–30 Oct (12.2, 12.1, most of 11.2, 11.1.7, most of 10.2). Panorama and WF-500/WF-500-B still have no fix date. If your branch has a released build, upgrade and then re-check management-interface exposure; if it is still pending, keep management access restricted to trusted sources. Patching web servers does not fix the appliance — PAN-OS ships its own embedded PHP.

2026-10-26 · [ASSESSMENT] — Found an independent government vendor-status list for embedded PHP: the Dutch CERT (NCSC-NL) maintains a TLP:CLEAR repo updated hourly with per-product status for both CVEs — supplier, product, versions, Vulnerable/Fix/Not vuln, and vendor-advisory links. It fills the gap our distro patch matrix can't: commercial appliances with embedded PHP. Its Palo Alto entry (PAN-OS and Panorama vulnerable, fixes not yet announced) independently matches our own PAN-OS assessment. Added to the package as the lookup layer for "is product X affected?" questions.

2026-10-26 · [ASSESSMENT] — Correcting yesterday's entry: the news article was never updated. The byte change I flagged came from the reader proxy we use to reach the Cloudflare-protected page, not from an edit — a normalised text comparison across three captures is identical. The substantive point stands unchanged: that article still carries the Unit 42 claim Unit 42 retracted, with no correction notice. The detection for its p2s-rider artefact stays (it was in the original text; we had simply not extracted it).

2026-10-25 · [ASSESSMENT] — The news article you sent has been updated but not corrected — its two new section headings mask that the Unit 42 claim it is built on was retracted. Meanwhile its update carries one genuinely new artefact: probes aimed at PAN-OS GlobalProtect honeypots carrying a PHP_SESSION_UPLOAD_PROGRESS marker p2s-rider (from 204.76.203[.]7), which reads as target-list building against appliances rather than generic scanning. Added a detection for the marker and recorded the artefact; ReliaQuest's lab sequence (POST → 500 → memory exhaustion → worker crash) also added to hunting notes.

2026-10-25 · [INCIDENT] — RETRACTION: Unit 42 withdrew its "European government intrusion" claim — and we are correcting our own correction. Unit 42 now states it "cannot confirm that initial access is associated with the specific PHP2Shell CVEs" and has retracted the 23 October update. What remains true: a real attack on a European government agency took place in the context of a vulnerable PHP application — but the CVE attribution is unconfirmed. So the position is back to no publicly confirmed victim tied to these two CVEs (CISA KEV still does not list them). Our earlier entry promoting this to a confirmed victim was premature: a single-vendor, uncorroborated victim claim should have been labelled vendor-reported and unverified. The 22.6 million exposed-instance estimate and all remediation guidance are unchanged.

2026-10-25 · [INCIDENT] — A confirmed intrusion now exists — and we are correcting an earlier statement. Palo Alto Networks Unit 42 (23 Oct) documented an attacker exploiting these CVEs against an unnamed European government entity, reaching password hashes and MFA secrets within two hours, building an NTLM-relay tunnel into Active Directory and attempting Domain Admin escalation. Command patterns (machine-readable ==SHELLUSER== / ==PASSWD== batches) point to agentic-AI post-exploitation. Separately, CYFIRMA describes a suspected DPRK/Lazarus-ecosystem campaign probing 216 hosts low-and-slow from 21 compromised nodes (moderate confidence). Our earlier line that "no confirmed victim exists" was wrong from 23 October — we had not monitored that source; it is monitored now. Remediation is unchanged.

2026-10-25 · [INCIDENT] — EPSS moved to near-certain exploitation. FIRST's daily update put CVE-2026-109618 at 0.99981 and CVE-2026-109619 at 0.99997 (from 11.8% and 6.2% the previous day). Verified against the official daily datasets across three days: same model version and control CVEs unchanged, so this is a real surge in exploitation evidence, not a re-scoring artefact. CISA KEV still does not list either CVE. Read it as exploitation intensity, not a per-host verdict.

2026-10-25 · [ASSESSMENT] — Verified a WAF bypass technique published by new research: PHP accepts Content-Type: multipart/form-data, boundary=X (comma instead of semicolon) and processes the upload normally, so rules anchored on the literal multipart/form-data; miss it. Reproduced on a stock PHP 8.4.26 image. Our detection rules are not affected — they match the substring without requiring the semicolon.

2026-10-24 · [ASSESSMENT] — Verified the two real PHP bugs behind the "patch is incomplete" claims. php-src#24481 is an open unserialize() use-after-free with no CVE — real, but not a bypass of the php2shell fixes. php-src#24488 is a regression introduced by 8.4.27 / 8.5.12: the new garbage collector double-frees SimpleXML state and crashes the worker (zend_mm_heap corrupted) on common patterns such as json_decode(json_encode($xml), true). Availability impact only; fix merged upstream but not yet released. Stage the upgrade and know the crash signature.

2026-10-24 · [ASSESSMENT] — Reviewed the community mitigation extension hp_uploadprogress_guard (143★): a native PHP module that mirrors the official fix on unpatched hosts and forces session upload progress off. Useful where patching is genuinely impossible, but it is a third-party native extension — supply-chain and stability risk, and not a substitute for the official fixed builds.

2026-10-24 · [ASSESSMENT] — Reproduced the public exploit: pre-auth remote code execution with error display disabled against stock PHP 8.4.26, on x86_64 and arm64. The deciding condition is Apache mod_php running a single prefork worker, not the architecture; multi-worker pools failed at the memory-leak stage. This disproved the widespread assumption that public tooling could not defeat display_errors=Off.

2026-10-24 · [PUBLISHED] — Detection package v2026-10-24 expanded and released: Suricata (18 rules), Zeek, YARA (8 rules), Sigma and ModSecurity, plus hunting queries for the public tool's *.p2s.php artefacts. Load-validated, versioned and downloadable with checksums from GitHub.

2026-10-24 · [PUBLISHED] — Live status page launched. Regenerated hourly from automated monitoring so the published position cannot drift from what is actually observed.

2026-10-24 · [PUBLISHED] — Client advisory PT-TIR-2026-0001 issued (single-file HTML, offline and print/PDF ready, TLP:AMBER). Available to clients on request.

2026-10-24 · [INCIDENT] — Palo Alto expanded PAN-OS Threat Prevention coverage from 5 signatures to 25 (content 9162–9164+). Estates that enabled only the original five are under-covered. Fixes remain pending (ETAs 26–30 October).

2026-10-23 · [ASSESSMENT] — Verified the widely-shared "second vulnerability / broken patch" claim against primary sources: no bypass of the critical fixes exists. The separate narrow advisory CVE-2026-115560 was documented with its exact preconditions and mitigation.

2026-10-23 · [PUBLISHED] — Package revised: PAN-OS exposure guidance, patch-availability matrix (distribution by distribution) and the clarification note on the second-vulnerability claim.

2026-10-23 · [INCIDENT] — Palo Alto confirmed PAN-OS and Panorama are affected, no special configuration required. Scope is the management web interface (GUI and XML API) and dataplane interfaces carrying a management profile. No fixes released; vendor assesses availability impact and says it is unaware of exploitation against its products.

2026-10-23 · [INCIDENT] — PHP published an informational advisory for a third, narrow CVE, CVE-2026-115560 (session upload progress). It requires an application that switches session.serialize_handler at runtime while upload progress is enabled; PHP describes it as an unlikely vector. Mitigation: session.upload_progress.enabled = 0. No patch yet — queued for the next regular release.

2026-10-22 · [ASSESSMENT] — Reviewed vendor and researcher statements on exploitability and ASLR bypass, and tested the loudest public claim against the exact image it referenced. Recorded what did and did not work, with the caveat that absence of a working public exploit is not a mitigation.

2026-10-22 · [PUBLISHED] — Detection package first circulated to clients: threat-intelligence dossier, load-validated detection rules, hunting queries and SOC triage guidance.

2026-10-22 · [INCIDENT] — The PHP security team publicly refuted claims that the released fixes were incomplete or bypassable, describing the reported behaviour as an unrelated issue with minimal impact that has nothing to do with the RCE. No re-spin of any fixed release has been published.

2026-10-21 · [INCIDENT] — Fixed releases published: PHP 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35 (14:00 UTC). Distributions followed within hours: Ubuntu USN-8948-1/-2, Debian 13 DSA-6574-1 and Debian 12 DLA-4848-1; official Docker php tags rebuilt the same day.

2026-10-21 · [INCIDENT] — Public disclosure. Advisories published as GHSA-w8hw-w9fx-j79w (CVE-2026-109618) and GHSA-6wmf-4rv5-p3rp (CVE-2026-109619), covering PHP 5.4.0 through 8.5.11 across all server APIs.

2026-10-17 · [INCIDENT] — Earliest observed in-the-wild exploitation attempts, four days before public disclosure, reported by Amazon's MadPot sensor network. Any compromise hunt should start from this date.

Full release notes, checksums and previous versions: GitHub releases.

Are you patched?

Patched means the fixed runtime is installed AND the PHP workers have been restarted. An upgraded package with stale workers still executes vulnerable code — this is the most common false sense of security in this incident.

PlatformFixedAdvisory / notes
PHP 8.58.5.12Upstream release
PHP 8.48.4.27Upstream release
PHP 8.38.3.36Upstream release
PHP 8.28.2.35Upstream release (oldest supported branch)
PHP 7.4 / 8.0 / 8.1source backports onlyNo official builds; use a distribution backport
Ubuntu 24.048.3.6-0ubuntu0.24.04.13USN-8948-1 / USN-8948-2
Ubuntu 22.048.1.2-1ubuntu2.28USN-8948-1 / USN-8948-2
Ubuntu 20.04–14.047.4.3-4ubuntu2.29+esm5 and older ESM buildsUSN-8948-1 / -2, including ESM
Debian 13 trixie8.4.27-1~deb13u1DSA-6574-1
Debian 12 bookworm8.2.35-1~deb12u1DLA-4848-1 — shipped as a DLA, not a DSA
Docker php images8.5.12 / 8.4.27 / 8.3.36 / 8.2.35 tagsPublished 2026-10-21
RHEL family / Remiverify with vendorNot independently confirmed
PAN-OS / PanoramapendingETAs 26–30 October; Panorama and WF-500 undated

Patching the application layer (WordPress, Drupal, Laravel, cPanel) does not address this. The runtime or appliance vendor must ship the fix.

Two practical traps: Debian 12's fix arrived as a DLA, so feeds filtered to DSA only will still show it as unpatched; and containers built before 21 October need rebuilding, not just restarting.

If you run Palo Alto PAN-OS or Panorama

Affected: PAN-OS 10.2, 11.1, 11.2, 12.1, 12.2; Panorama; WildFire WF-500 / WF-500-B.

Not affected: Cloud NGFW, Prisma Access and Prisma Cloud, GlobalProtect App, Cortex XDR/XSIAM/XSOAR/Xpanse, Bridgecrew. Cortex XCORE (21 Oct) and WildFire Cloud (22 Oct) were already remediated.

Until fixes ship: keep the management interface off untrusted networks, and enable the current Threat Prevention ID set. As of 24 October that is 25 signatures — 543004, 97622–97625, 97643, 97647, 97649, 97652, 97653, 97656–97659, 97662–97664, 97670–97674, 97676–97678 — across content versions 9162, 9163, 9164 and later.

Those signatures only protect the management plane if all of the following are true: management traffic is routed through a dataplane port, the inbound management certificate has been replaced, inbound traffic to management is decrypted, and threat prevention is enabled on management services. Without that chain the signatures do not see the management plane.

Note that an appliance's embedded PHP cannot be patched by you — only PAN-OS maintenance releases (or the vendor, for rebadged appliances such as Siemens RUGGEDCOM APE1808 per SSA-566512) can address it. Siemens also lists a wider affected software portfolio in SSA-166980 (Spectrum Power, Energy Engage, blueplanet NX3, AMIS TS), most of it with no fix available.

Detection

Because the vulnerable code runs before application code, application logs are frequently empty. Behavioural signals are more reliable than signature matching:

  • Oversized PHPSESSID response cookie — the legitimate value is 26 characters and never percent-encoded; a multi-kilobyte value indicates memory disclosure. This is the single highest-value alert, and it indicates credential exposure even when exploitation fails.
  • Multipart bodies far larger than the application accepts, particularly without a Content-Length header (chunked encoding bypasses body-size checks).
  • PHP worker restarts or 502 bursts correlated with large multipart uploads.
  • Oversized session files in shared session directories.

An oversized response cookie should be treated as a credential-rotation event, not a failed attack: memory disclosure means other requests' secrets may have been read.

Scope details that are widely misreported

  • PHP 7.4, 8.0 and 8.1 are affected, despite advisory titles naming 8.2–8.5. Upstream published source-only backports for those branches; distributions shipped builds.
  • display_errors = Off is not a security control. Private research has produced working chains with ASLR enabled and error display disabled, and a public PHP-FPM chain exists for specific multi-pool configurations. The practical barrier is exploit maturity, not configuration hardening.
  • The silent disclosure path is the underestimated half. Memory can be read without code execution, so absence of a shell does not mean absence of compromise.
  • Confirmed public capability changes the calculus. As of 24 October a public tool achieves code execution with error display disabled on single-worker Apache mod_php/prefork. FPM deployments are unaffected by that tool; worker count is not a mitigation.
  • Concrete artefact from that tool: it plants a webshell at /.p2s.php (also /assets/.p2s.php) and executes commands via ?c=, removing itself via ?rm. Any request to a *.p2s.php path is worth alerting on, and the file name is a reliable indicator even where the exploit itself failed.
  • Exploitation attempts began before public disclosure. Any hunt should start from 17 October 2026, not from patch day.

What to do today

  1. Confirm the PHP version actually serving requests, per host and per pool — not the package version on disk. Restart workers after upgrading.
  2. Patch what you can reach and isolate what you cannot (end-of-life PHP, vendor appliances).
  3. Set session.upload_progress.enabled = 0 where applications do not need it. This is PHP's own mitigation for the session/upload-progress path.
  4. If you run multi-pool PHP-FPM with pooled session storage and mismatched memory limits, review that configuration — it is the deployment shape public FPM chains target.
  5. Hunt from 17 October 2026, and treat any oversized PHPSESSID cookie as credential exposure.
  6. For PAN-OS, treat management-plane exposure as the deciding factor and confirm the full current Threat Prevention ID list is enabled.

Sources

  • PHP release announcement and security advisories — php.net/archive
  • GitHub Security Advisories: GHSA-w8hw-w9fx-j79w (CVE-2026-109618), GHSA-6wmf-4rv5-p3rp (CVE-2026-109619)
  • Ubuntu: USN-8948-1 and USN-8948-2
  • Debian: DSA-6574-1 (trixie) and DLA-4848-1 (bookworm)
  • Palo Alto Networks advisory for CVE-2026-109618 (updated 2026-10-24)
  • Amazon Threat Intelligence, MadPot pre-disclosure exploitation report (2026-10-21)
  • oss-security mailing list threads, 21–23 October 2026

Maintained automatically from a live monitoring pipeline. Detection package referenced above: php2shell-detection-package-2026-10-26 (sha256 124977220781739c…). No exploit code is published on this page.