daily cyber × ai intelligence

← index

tagged

[exploit-chain]

2 editions

November 1, 2026 weekly

PHP2Shell Reached Root as Models Closed the Patch Gap

PHP2Shell reached production with Fragnesia kernel exploitation, rooting 25 servers across the US, Japan, Taiwan and South Korea including a US government Active Directory environment. GLM 5.3 autonomously built complete Firefox exploit chains from patch information without human input, while poisoned Claude Code extensions in Cursor marketplace and GitLab CVE-2026-97229 exploits demonstrated how developer tooling became privileged attack infrastructure. Management-plane attacks deepened across Cisco Catalyst SD-WAN, Tailscale SSH, and Azure API Connections, while ad networks from Discover-Tech supplied spyware delivery across Finnish critical infrastructure.

October 25, 2026 weekly

PHP2Shell’s Noisy Exploit Race Met Agentic Operations at Scale

PHP2Shell exploitations escalated from pre-disclosure scanning to confirmed post-exploitation with environment sensitivity, while Storm-0940 operationalized ARTEX AI-assisted intrusions across 2,250+ Microsoft tenants using agentic workflows. LiteLLM suffered a cascading supply-chain compromise starting from an unrevoked master key that enabled GitHub token theft and poisoned package releases. Durable credentials including OAuth tokens, rogue Entra devices, and embedded Brotli decoder vulnerabilities across Node.js, Python, Bun, and multiple other ecosystems replaced password-centric threat models, while privilege-escalation paths in management planes (VeloCloud Orchestrator, FortiWeb, PAN-OS) remained exploitable despite patching.