Black October: AI Agents in Attacks, PHP Vulnerability, and...

Black October: AI Agents in Attacks, PHP Vulnerability, and Compromised LiteLLM

On October 21–22, 2026, four incidents at the intersection of AI and infrastructure accumulated: CrowdStrike reported on groups attacking with AI agents, a critical 2004 vulnerability was discovered in PHP, a LiteLLM 1.107.0 release with malicious code was available on PyPI for about two hours, and Railway reported possible third-party access to client secrets.

Compiled by Sergey KostenchukPublished 2026-10-22Updated 2026-10-22

2026-10-22 Safety OpenAI

🛡 Black October: AI Agents in Attacks, Vulnerable PHP, and LiteLLM

CrowdStrike described four groups of attackers who, from May to August, used AI agents (including Claude Code); the victims were 15+ medical organizations in Argentina. Then a critical 2004 vulnerability was found in PHP, a LiteLLM release was compromised, and Railway reported possible access to client secrets.

🌍 According to CrowdStrike, AI agents autonomously traverse the kill chain in approximately one hour — the report lacks methodology, serving as a signal of the threat direction. GPT-5.6 Sol helped find PHP vulnerabilities that had existed for 22 years.

👤 LiteLLM: if 1.107.0 is installed within the window of 11:39–13:23 UTC on October 21 — switch to 1.106.2 (credential theft, RCE). Railway — urgently rotate API keys and secrets. PHP — update to 8.5.12.

Source 1: https://journal-production-3e67.up.railway.app/articles/black-october Source 2: https://www.crowdstrike.com/en-us/blog/crowdstrike-uncovers-new-agentic-ai-campaigns/

Sources