Tuesday, November 17, 2026
Paperclip Index
Paperclip IndexControl failures4Control failures onlyA week ago: 9, negligible harmThe Index

SignalCapabilitySG-0048

CISA confirmed attacks on the two critical PHP flaws AI helped find, after attempts began before disclosure

Amazon's decoy servers logged an exploit attempt on Oct. 17, four days before PHP fixed the two bugs that let unauthenticated attackers run code. Broad attempts followed the release. CISA has added both flaws to its list of exploited vulnerabilities, giving federal agencies until Oct. 29 to patch, and a Perplexity researcher logged 35 working shells on test servers set up in a way that favored the attack. Calif, OpenAI and Wiz reported the flaws; Wiz credits Google's Gemini 4 Argon and OpenAI's GPT-5.6 Sol with much of its work.

Why it matters

Amazon says AI-assisted analysis makes it cheaper to turn a patch into a working exploit, so more attackers can act within hours, and here attempts began before the planned disclosure. The PHP Foundation, which scanned PHP with Anthropic's Claude Mythos earlier this year, says tools now find bugs faster for attackers and defenders alike, and its small team has handled more than 100 invalid, duplicate or irreproducible reports since July 1.

Sources

Read the reporting

9 sources. Links go to the original publishers; the summary above is in our own words.

  1. primaryPatch the Planet: global response coordination as agents uncover critical vulnerabilitiesOpenAI · Oct. 21, 2026openai.com/index/coordinated-disclosure-of-critical-vulnerabilities
  2. primaryRCE via OOM corruption (GHSA-w8hw-w9fx-j79w)PHP (GitHub security advisory) · Oct. 21, 2026github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w
  3. primaryAmazon Threat Intelligence identifies PHP2Shell exploitation prior to public disclosure in MadPotAmazon Web Services · Oct. 21, 2026aws.amazon.com/blogs/security/amazon-threat-intelligence-identifies-php2shell-e…
  4. primaryToday's PHP Security Release, and How We're Strengthening PHP SecurityThe PHP Foundation · Oct. 21, 2026thephp.foundation/blog/2026/10/21/todays-php-security-release-and-how-were-stre…
  5. primaryPHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the InternetWiz · Oct. 21, 2026wiz.io/blog/php2shell-php-rce
  6. primaryPHP: News Archive - 2026PHP · Oct. 21, 2026php.net/archive/2026
  7. researchphp 2 shell: A tale of two PHP remote vulnerabilitiesCalif · Oct. 21, 2026calif.io/php2shell/
  8. newsCISA confirms PHP2Shell under attack as researchers clock successful exploitsThe Register · Oct. 27, 2026theregister.com/cyber-crime/2026/10/27/cisa-confirms-php2shell-under-attack-as…
  9. blogPHP2Shell: Critical PHP RCE (CVE-2026-109618, CVE-2026-109619)Sansec · Oct. 21, 2026sansec.io/research/php2shell

Signals

More from the news desk

  1. Capability

    METR found Claude Opus 5.5 could pick out every alignment test from ordinary tasks in its set

  2. Capability

    Google DeepMind's AI agents formally verified Signal's encryption code down to machine code in four weeks

  3. Capability

    Attackers ran AntiFox, an open-source AI pentesting agent forked from ARTEX, against web servers and AWS accounts

All signals

Cite and share

Use this signal

Citation

Paperclip Index. “CISA confirmed attacks on the two critical PHP flaws AI helped find, after attempts began before disclosure.” Signal SG-0048. Reported Oct. 21, 2026; updated Oct. 29, 2026. https://paperclipindex.com/signal/SG-0048