PHP zero-day scramble and Microsoft's disclosure of agentic AI attacks mark worst cybersecurity day

October 22 was dubbed by TheZvi as "possibly the worst cybersecurity day ever": on one side, two critical PHP remote code execution vulnerabilities were hastily patched after coordinated disclosure broke down; on the other, Microsoft revealed that a state-backed hacker group had used an agentic AI framework to run automated attacks against hundreds of organizations worldwide, with several cases gaining privileged access to cloud infrastructure within 3 hours. Together, many commentators saw this as a landmark moment signaling "AI attacks have entered the real battlefield."

Confirmed

• PHP released emergency updates across versions 8.5.12, 8.4.27, 8.3.36, and 8.2.35, fixing two related critical unauthenticated RCE vulnerabilities, CVE-2026-109618 and CVE-2026-109619; when conditions are met, attackers can execute code on servers without logging in—officials urge immediate upgrades
• Vulnerability details leaked before the patch release, breaking the coordinated disclosure process and forcing security teams to ship fixes early; since examining the fix reveals the flaw, the industry warned of mass attack attempts within about 15 minutes of the fixed versions going public, with millions of php2shell-related exploit attempts
• The FBI's cyber division issued a global alert, and the NSA urgently notified all PHP users to prepare for updates; if immediate updating isn't possible, systems should be taken offline until fully mitigated
• Microsoft threat intelligence report: since September 2026, the China-linked group Storm-0940 has used ARTEX, an open-source agentic AI penetration testing framework on GitHub, chaining LLM agents into a full pipeline from reconnaissance to data exfiltration with almost no human intervention, automatically attacking hundreds of organizations across multiple industries; in several cases, going from initial targeting to privileged cloud infrastructure access took under 3 hours, hitting over 2,250 cloud tenants; Microsoft also attributed the recently disclosed ConnectWish breach—ConnectWise intrusion—to the same group
• Per Hesamation's relay of the PHP Foundation security update, Anthropic's Claude (Mythos) scanned PHP codebases earlier this year, and the exploit could be verified using OpenAI's Daybreak Red

Unconfirmed

• Impact figures vary across posts: some say 70% of websites use PHP, others say 80% of websites are affected with CVSS 9.2; the exact figures haven't been reconciled
• @kimmonismus and @LinusEkenstam called this AI-driven attack wave "possibly far more serious than the Hugging Face incident"—an insider assessment, not an official conclusion

Why it matters

• @teortaxesTex quipped sarcastically that if autonomous agents didn't exploit the PHP zero-days to take down a third of the internet with zero barriers, p(doom) should be revised down—underscoring that attack barriers are now low enough for it to happen anytime
• Ethan Mollick, quoted by @eldonredwards, called this a visible signal of what the AI community has warned about since Mythos launched: offense and defense accelerating each other at inhuman speed, while the human systems that must respond still move at human speed
• @LinusEkenstam and others stressed that AI-driven attacks already have real victims; ARTEX proves open-source LLM agents can form end-to-end attack chains, putting fundamental pressure on traditional human-speed security operations

2026-10-22 ~ 2026-10-23 · 24 related posts

Full story(15 episodes)→

• Episode 1: PHP Rushes Emergency Patches for Two Unauthenticated RCE Flaws Found with OpenAI Models(2026-10-21, 113 posts)
• Episode 2: Researchers Predict LLM Audits Will Uncover More PHP Vulnerabilities(2026-10-21, 3 posts)
• Episode 3: PHP RCE proof-of-concept confirmed working, but not one-click(2026-10-21, 2 posts)
• Episode 4: Black October: 27 Chained Security Incidents Hit Open Source and Cloud Infrastructure(2026-10-21, 21 posts)
• Episode 5: LLM-Generated Fake PHP Exploit Sparks Security Panic(2026-10-21, 2 posts)
• Episode 6: Open-source repo demonstrates HTTP-only ASLR bypass RCE in PHP 8.4.26(2026-10-21, 2 posts)
• Episode 7: PHP 8.5.12 Patch Incomplete: php2shell Double-Free Still Exploitable(2026-10-22, 2 posts)
• Episode 8: php2shell Vulnerability Story Told in Hand-Drawn Comics(2026-10-22, 2 posts)
• Episode 9: PHP zero-day scramble and Microsoft's disclosure of agentic AI attacks mark worst cybersecurity day(2026-10-22, 24 posts)
• Episode 10: Critical PHP RCE Flaws Leaked Before Patch, Scans Surge Within 20 Minutes(2026-10-22, 3 posts)
• Episode 11: PHP2Shell Exploited Four Days Before Public Disclosure(2026-10-22, 2 posts)
• Episode 12: php2shell Exploits Top One Million After Disclosure(2026-10-22, 3 posts)
• Episode 13: Drupal/PHP Vulnerability Surge Keeps Security Teams Busy Hardening Sites(2026-10-22, 2 posts)
• Episode 14: Investor Deedy Warns AI Is Lowering the Barrier to Exploiting Vulnerabilities(2026-10-22, 3 posts)
• Episode 15: Researcher pushes back on php2shell panic as AI agents clean up(2026-10-23, 2 posts)

• NSA urges PHP users to patch two critical vulnerabilities or go offline — AnnaCiaunica · 2026-10-22
• Microsoft: Storm-0940 used agentic AI framework ARTEX to attack hundreds of orgs — basedjensen · 2026-10-22
• Critical PHP zero-day (CVSS 9.2) disclosure sparks AI-powered patch-race warnings and local-agent quips — AIFlow_ML · 2026-10-22
• NSA issues emergency patch-or-go-dark advisory for PHP flaw powering 43% of the web — churchkey · 2026-10-22
• Microsoft: China-linked Storm-0940 used agentic AI framework ARTEX, breaching clouds in under 3 hours — scottleibrand · 2026-10-22
• Leaked PHP vulnerability forces early patch, massive attacks expected within 15 minutes — i_dg23 · 2026-10-22
• PHP critical RCE bugs leak pre-patch; agencies urge servers offline, 1.5M unpatched — eldonredwards · 2026-10-22
• FBI issues global alert on two critical PHP RCE vulnerabilities, advises taking unpatched systems offline — denisparra · 2026-10-22
• AI circle quips: update your p(doom) if agents didn't exploit the PHP emudhra bug — teortaxesTex · 2026-10-22
• Microsoft: China-linked Storm-0940 used open-source ARTEX AI agent to breach clouds in under 3 hours — rohanpaul_ai · 2026-10-22
• Microsoft details ARTEX AI-agent nation-state supply chain campaign and mitigations — rohanpaul_ai · 2026-10-22
• PHP2Shell: CVSS 9.2 PHP flaws knock governments offline as AI compresses exploit dev from months to a week — APPSO · 2026-10-22
• AI-powered attacks hit hundreds of orgs, gaining privileged cloud access in under 3 hours — kimmonismus · 2026-10-22
• PHP ships emergency fixes for two unauthenticated RCE bugs after embargo break — kingslyj · 2026-10-22
• [source] PHP Foundation reveals Claude scanned PHP code, frontiers models helped find two critical RCE bugs — Hesamation · 2026-10-22
• Unpatched PHP flaw lets attackers take over servers, leaving ~70% of the web exposed — josh_bickett · 2026-10-22
• AI Cyber Arms Race Escalates: China-Linked Hackers Use Agents to Breach 2,250+ Cloud Tenants — 141_1337 · 2026-10-22
• [source] Microsoft: China-linked Storm-0940 used agentic AI framework ARTEX to attack hundreds of orgs — testingcatalog · 2026-10-22
• Microsoft: AI-Powered Attacks Hit Hundreds of Orgs, Cloud Privileges in Under 3 Hours — LinusEkenstam · 2026-10-22
• [source] TheZvi newsletter: PHP zero-day marks worst day in cybersecurity history; Anthropic's draft bill leak deepens — TheZvi · 2026-10-22
• AI agent swarm hit 400 Latin American banks amid 48-hour exploit storm — 新智元 · 2026-10-23
• AI-found PHP flaws put up to 70% of websites at risk, triggering a global patch race — LouMM · 2026-10-23
• PHP 9.2 Flaws, Poisoned AI Tool with 23M Downloads, Rogue AI Agents in Major Banks—All in 48 Hours — r0ck3t23 · 2026-10-23

1 near-duplicate retellings: CurieuxExplorer