PHP2Shell: critical remote-code-execution flaws in PHP, found with GPT-5.6 Sol and Gemini 4 Argon, force emergency patching and take websites offline worldwide
Confirmed
Importance: historic (5 of 5)The takeaway
The bugs were found with frontier AI: by Calif with GPT-5.6 Sol under OpenAI’s Patch the Planet programme, and by Google’s Wiz with Gemini 4 Argon, GPT-5.6 Sol and its Wiz Atlas agent.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- Historic (5 of 5)
- Last verified
- 25 October 2026
Your AI and this story
- GPT-6 Astra174 days after its cutoff
- Claude Opus 5.5113 days after its cutoff
- Gemini 4 Argon204 days after its cutoff
- Grok 4.7143 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 113 days before it.
Key facts
- CVEs: CVE-2026-109618 and CVE-2026-109619, severity 9.2/10, rated critical by PHP (The Record); Wiz describes four bugs under the two CVEs: an upload-progress unserializer use-after-free, remote process-memory disclosure, a fatal-error backtrace use-after-free and OPcache type confusion
- Scope (Wiz): ‘Every PHP version released in the last decade is affected by at least one of these vulnerabilities, going back to version 5.4’; TechCrunch says the flaws go back to PHP 5 (2004). PHP 8.1 and older get no fixes; upgrade target 8.5.12 (TechCrunch)
- Impact: unauthenticated RCE demonstrated against WordPress, Drupal and Joomla; memory disclosure can leak passwords, tokens and API keys; ‘Exploitation does not depend on a flaw in the application’s code’ (Wiz)
- Wiz timeline: first bug found Aug 19, 2026; RCE on Drupal Aug 20 and WordPress Aug 21; reported to PHP Aug 24; PHP acknowledged Sept 21; patched releases and disclosure Oct 21
- Who found it: Wiz Research (Sagi Tzadik, Shir Tamari, Nir Ohfeld, Tomer Schwartz) with Wiz Atlas, Gemini 4 Argon and GPT-5.6 Sol; Gemini 4 Argon ‘made an invaluable contribution to finding the vulnerability, exploit development, attack detection, and validation of the patch’. Separately, security firm Calif found a flaw with GPT-5.6 Sol under OpenAI’s Patch the Planet (TechCrunch, The Record)
- Wiz: ‘a single researcher completed in a week what previously would have taken a world-class team months of work’
- OpenAI (X, 5.3M views): ‘we consider this bug to be one of the most impactful vulnerabilities in recent memory. It also marks a new phase, where AI can find security bugs with internet-scale impact’; it led a two-month coalition of maintainers and security providers because ‘the fix was quite intricate and required a major change to the PHP engine’
- OpenAI released a Codex skill to check whether a site is affected (github.com/openai/php2shell-checker, 1.2M-view post)
Show 10 more
- Fallout (The Record): PHP pre-announced the bugs on the evening of Oct 20 and patches landed after 10 a.m. ET Oct 21; the EU, US, Germany, Japan, Australia, the UK and the Netherlands issued warnings; several governments and companies such as Nintendo took sites offline; hundreds of sites were down on Oct 21
- NSA advised that sites that cannot be patched immediately be taken offline (TechCrunch)
- Exploitation: Cloudflare CTO Dane Knecht said attacks were already emerging; watchTowr saw in-the-wild attempts ‘with a variety of proof-of-concept tools vibecoded with the help of AI’, none successful yet (The Record). OpenAI and Wiz said they knew of no exploitation before disclosure
- Interim mitigations from Cloudflare, Fastly, Google, Vercel, Akamai, AWS and others; Wiz: ‘Applying the PHP security update remains the only complete remediation’
- Four-day head start (Amazon Threat Intelligence, Oct 22): its MadPot honeypot network saw a PHP2Shell exploitation attempt on Oct 17, four days before disclosure, from an IP tied to a commercial anonymisation service; Amazon says it is unclear how the actor got exploit information before disclosure, and whether the attempt would have worked against real systems. Amazon also saw multiple proof-of-concept exploits on Oct 21, the earliest about 02:00 UTC, ~12 hours before the 14:00 UTC embargo end, with varying payloads from different infrastructure, suggesting several actors
- Exploit volume (Cloudflare CEO Matthew Prince, Oct 22, 1.86M views): ‘In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.’
- Cloudflare says it deployed WAF mitigations by default for all free and paid customers about nine hours before public disclosure (blog.cloudflare.com/php-vulnerabilities-waf-protection, via Techmeme)
- Agency and vendor advisories: CISA alert ‘Critical Vulnerabilities in PHP Require Immediate Patching’ (Oct 21, revised Oct 23): ‘CISA strongly urges organizations to immediately patch’; inventory PHP including embedded copies, take systems offline if they cannot be updated or run PHP 8.1 or earlier, use a WAF only as a stopgap; no exploitation observed by CISA at publication; the Oct 23 revision adds a SIGMA detection rule. CERT Polska advisory 196/2026 (Oct 21): versions from PHP 5.4 are affected, interim php.ini settings (memory_limit, post_max_size ≤ 8M, max_input_vars ≤ 1000, session.upload_progress.enabled off) only reduce risk: ‘Jedyną skuteczną metodą ochrony … jest instalacja poprawionej wersji PHP’ (the only effective protection is installing a fixed PHP). Sophos (Oct 21–22): no Sophos products affected; its legacy MyUTM licensing portal ran an affected PHP and was taken offline on Oct 21 to update, with no sign of exploitation
- Agentic post-exploitation (Unit 42 threat brief, update of Oct 23): attackers used ‘agentic AI attack platforms’ after exploiting PHP2Shell; against a government entity in Europe they reached password hashes and MFA secrets within two hours of initial exploitation, then built an NTLM-relay tunnel into Active Directory and attempted Domain Admin escalation (outcome not stated). Unit 42 counted over 22.6M exposed, potentially vulnerable instances (Cortex Xpanse). See Unit 42: attackers used an agentic AI attack platform after PHP2Shell exploitation, reaching a European government entity’s password hashes and MFA secrets within two hours
- Patch gap: the PHP Foundation’s patches cover PHP 8.2 and newer, leaving legacy deployments dependent on third-party vendors (Wired via Techmeme); fixed versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35, with backports reported for 7.4–8.1
What happened
PHP’s security team warned on the evening of Oct 20 of two critical bugs and released fixed versions on Oct 21. The same day Wiz (owned by Google) published its research and OpenAI said its models had helped find the flaws through Patch the Planet, its programme that gives vetted defenders frontier models for vulnerability work. Both groups disclosed responsibly and spent about two months with PHP and infrastructure providers on a fix that, OpenAI said, “required a major change to the PHP engine”.
Once the bugs were public, a patching race began. Governments in Europe, the US, Japan and Australia issued alerts, the NSA told operators who could not patch to take sites offline, and some government sites and companies (Nintendo among them) went dark. Within hours Cloudflare and watchTowr reported attack attempts, some using proof-of-concept code written with AI help.
OpenAI’s announcement does not name every finder in the post we could read (the OpenAI page returned HTTP 403 to our fetcher); the attribution of one flaw to Calif with GPT-5.6 Sol comes from TechCrunch and The Record.
Why it matters
It is the first AI-found vulnerability with internet-scale reach to trigger a global emergency response, and OpenAI presents it as proof of its “Defender’s Window” argument: frontier models found the bug before attackers did. Wiz’s note that AI removes the old barrier to exploitation (“With AI this is no longer a barrier”) shows the other side: the same models shorten the time from disclosure to working exploit, so millions of unpatched or unpatchable PHP servers are now exposed.
Sources
13 sources from 12 sites. Numbers match the chips in the text.
13 sources: 10 primary, 2 press, 1 reaction
Primary
- Wiz Research: PHP2Shell — AI uncovers critical RCE vulnerabilities affecting 70% of the Internetwiz.io, official
- OpenAI: Patch the Planet — global response coordination as agents uncover critical vulnerabilitiesopenai.com, official
- OpenAI on X (5.3M views)x.com, official
- GitHub: openai/php2shell-checker (Codex skill)github.com, code
- AWS Security Blog: Amazon Threat Intelligence identifies PHP2Shell exploitation prior to public disclosure in MadPotaws.amazon.com, official
- CISA: Critical Vulnerabilities in PHP Require Immediate Patching (Oct 21, revised Oct 23)cisa.gov, official
- CERT Polska 196/2026: Krytyczne podatności w PHP umożliwiające zdalne wykonanie kodu (php2shell)moje.cert.pl, official
- Sophos advisory SA-20261021: PHP2Shell vulnerabilitiessophos.com, official
- Unit 42 (Palo Alto Networks): Threat Brief on CVE-2026-109618 and CVE-2026-109619 (updated Oct 23, agentic post-exploitation)unit42.paloaltonetworks.com, official
- Matthew Prince on X: exploit attempts before and after disclosurex.com, official
Press
- The Record: Hundreds of websites taken offline as concern grows over two PHP vulnerabilitiestherecord.media, press
- TechCrunch: PHP Armageddon — OpenAI, Google’s Wiz find ancient holes, panicking the internettechcrunch.com, press
Reactions
- Hacker News discussion (441 points)news.ycombinator.com, discussion
Changes
- Added CISA’s Oct 21 alert, CERT Polska advisory 196/2026, Sophos advisory and Unit 42’s threat brief with its report of agentic AI post-exploitation (own entry Unit 42: attackers used an agentic AI attack platform after PHP2Shell exploitation, reaching a European government entity’s password hashes and MFA secrets within two hours); items cited by Gemini, each verified on the primary page
- Added AWS/MadPot’s finding of an exploitation attempt on Oct 17, four days before disclosure, Cloudflare’s exploit-volume numbers from Matthew Prince, Cloudflare’s pre-disclosure WAF mitigations and the 8.2-and-newer patch limit. The AWS blog was found via search (the Wired and Ars stories are blocked for us).
- Added Wiz’s video with the finders Sagi Tzadik (Wiz) and Thai Duong (Calif).
- Filed (HN, lab X, Google News)