Security Week: PHP2Shell Scans, eMudhra Exit, and AI Agent Exploits

Key points

• PHP Foundation patched two critical unauthenticated RCE flaws (CVE-2026-109618/19), codenamed PHP2Shell, which Cloudflare reported triggered millions of exploitation attempts, though no successful breaches were confirmed.
• eMudhra exited the TLS certificate business after eight intermediate CAs were found to use weak private keys; Apple and Python packages updated their trust stores to block the affected certificates.
• A critical RCE flaw in Google’s Brotli decoder (CVE-2026-111497), dubbed Node2Shell, was patched across multiple libraries including Node.js and Python.
• Plex Media Server vulnerabilities were actively exploited in the wild to deploy cryptocurrency miners, leading CISA to add them to its Known Exploited Vulnerabilities catalog.
• LiteLLM suffered a supply chain attack after a master key was leaked to a public GitHub repository, allowing attackers to push a malicious package version that harvested credentials and installed backdoors.
• CrowdStrike identified four threat actors using Claude Code to automate offensive cyber operations, including breaches of healthcare and government entities.

Background

This week’s chaos follows a broader trend of AI-driven risks and supply chain disruptions noted in recent coverage. While Amazon faces logistical hurdles in its massive AI infrastructure push, and Axios warned of cascading societal disruptions from converging AI and cyber threats, the current security landscape is defined by the rapid weaponization of new vulnerabilities. The shift from AI drafting code to AI executing tactical decisions in attacks, as noted by Rapid7, aligns with the observed use of agentic AI by threat actors.

Why it matters

The convergence of unauthenticated RCE flaws in ubiquitous software and the rise of AI-driven attacks significantly reduces the window for defensive response. The compromise of trust anchors like eMudhra’s certificates and the exploitation of AI agent trust mechanisms (e.g., AGENTS.md files) indicate that attackers are targeting foundational layers of software and AI infrastructure, not just application-level bugs.

What to watch

Organizations must immediately patch PHP, Brotli, and Plex vulnerabilities, restrict unprivileged user namespaces for Linux KSMBD flaws, and revoke sensitive environment variables on platforms like Railway. Security teams should monitor for AI agent misuse, update CA bundles to exclude eMudhra certificates, and prepare for the rapid exploitation cycle of newly disclosed CVEs.