PHP2Shell: AI uncovers critical RCE in PHP
Points and comments are a snapshot, not live.
AI helped discover critical PHP remote code execution vulnerabilities affecting 70% of websites.
Wiz Research used frontier AI models (Gemini 4 Argon, GPT-5.6 Sol, Wiz Atlas) to find multiple memory-safety vulnerabilities in PHP core, grouped under CVE-2026-109618 and CVE-2026-109619 (CVSS 9.2). The flaws affect every PHP version since 5.4, allowing unauthenticated remote code execution and memory disclosure. Exploitation chains combine bugs in upload-progress unserializer, fatal-error backtrace, and OPcache. The researchers demonstrated pre-authentication RCE on WordPress, Drupal, and Joomla. PHP released a patch on October 21, 2026. The vulnerabilities had been present for over a decade; AI reduced the effort to develop working exploits from months to a week.
What commenters are saying
Commenters question whether AI truly found new bugs or rediscovered previously reported ones, citing a 2020 PHP bug report (#78876) for a similar OOM issue that was fixed incompletely. Some argue that decades-old code, reviewed by few, is exactly where AI can help. Others express skepticism about the novelty and hype. Several commenters criticize the timing of the disclosure as overly aggressive, released minutes after the fix. One commenter notes that Anthropic's Mythos had previously scanned PHP without finding this vulnerability, raising questions about model differences. The thread splits between AI-skeptical responses and those impressed by the discovery method.