Threat Briefing: October 23, 2026

This week showed software vulnerabilities, supply chain attacks, and AI security risks converging. Two critical PHP flaws (PHP2Shell) are under active exploitation, and a CVSS 10.0 GitLab bug allows unauthenticated remote code execution.

Attackers also compromised the LiteLLM AI gateway’s release pipeline to distribute credential-stealing malware. Certificate authority eMudhra is exiting the certificate business amid weak-key concerns, and Microsoft introduced controls that treat AI agents as potential insider risks.

Check out our insights about PHP2Shell and recommended actions to take immediately.

• Critical Vulnerabilities Expose Risks Across PHP Environments – Two critical PHP flaws, collectively called PHP2Shell, were disclosed on October 21 after AI-assisted code reviews found them in August. Emergency patches followed, and hundreds of websites went offline. Coordinated disclosure broke down when a researcher accidentally published a proof-of-concept video and Chainguard released its patch early, speeding up exploit development. The Record
• Microsoft Targets Emerging Threats from AI Agents – Microsoft announced Guardian Agents, Dynamic Permissions, and Dynamic Agent Risk as extensions of Project Perception, its agentic security system. Together, they adjust agent access and containment in real time based on risk signals, behavior, and organizational policy. Help Net Security
• New GitLab Flaw Exposes Self-Managed Instances to RCE Risk – GitLab patched CVE-2026-110264 on October 21. The CVSS 10.0 integer overflow in the bundled Onigmo regex compiler allows unauthenticated remote code execution on self-managed servers. It affects every release from version 7.0 (2014) onward, making it the broadest of four regex-engine flaws reported by the same researcher in one month. The Hacker News
• LiteLLM Supply Chain Attack Delivers Credential-Stealing Malware – On October 21, attackers compromised LiteLLM’s release pipeline and published malicious version 1.107.0 to PyPI and Docker registries. The artifacts carried credential-stealing malware and a remote code execution loader, and stayed live for about two hours. BleepingComputer
• eMudhra Withdraws After Critical Certificate Security Failure – Indian certificate authority eMudhra is exiting the global SSL/TLS certificate business after a researcher found that private keys for eight of its intermediate certificates weren’t randomly generated, one being simply the number four. The flaw dated to an April 2025 key ceremony and went unnoticed for 18 months through two audits. Firefox, Apple, and Chrome have blocked the affected certificates. The Record