Linux kernel MPTCP flaw enables remote root while PHP2Shell roots Palo Alto firewalls

A critical Linux kernel flaw (CVE-2026-98432) can let unauthenticated attackers run commands as root on servers with Multipath TCP listeners, and Anthropic researchers demonstrated a working exploit. Meanwhile, PHP2Shell attacks have rooted Palo Alto firewalls by chaining the PHP bugs with a kernel privilege-escalation flaw.

Security

Linux kernel MPTCP flaw allows remote root, and Anthropic built a working exploit

CVE-2026-98432 is a use-after-free race condition in the kernel’s IPv4 TCP code, scored 9.8 out of 10 and affecting kernels back to 4.4 across major distributions. Anthropic says an unreleased Claude model built a working exploit that reached remote root on a server running an MPTCP (Multipath TCP) listener. Fixes are available upstream and in Debian, but Ubuntu and RHEL releases were still waiting as of October 30. Note that Go 1.24 enables MPTCP by default on Linux, so some Go services may be exposed without their owners knowing. Disabling MPTCP alone is not enough, since the bug also triggers via ordinary TCP. CIQ says firewall rules blocking IPv4 TCP packets with IP options can help until patches arrive. No exploitation in the wild has been confirmed. Source

PHP2Shell attacks root Palo Alto firewalls but WordPress sites hold so far

A China-linked group chained the PHP2Shell flaws CVE-2026-109618 and CVE-2026-109619 with the Fragnesia kernel privilege-escalation bug to install webshells with root privileges on Palo Alto PAN-OS firewalls, compromising at least 25 devices. Palo Alto has patched all affected PAN-OS branches plus Panorama and urges customers to keep the management interface off the internet. Attacks on WordPress sites have failed so far, but Defiant stresses that updating PHP remains essential. Source

TeamViewer flaw let controlled machines attack the connecting computer

TeamViewer has fixed a flaw where a compromised host (the machine being controlled) could attack the viewer (the computer connecting to it), crashing its TeamViewer software and potentially more. Update to the latest version. Source

Chinese hackers use AI-powered ARTEX framework against 200+ organizations

CrowdStrike attributes attacks using ARTEX, a Go-based, AI-powered penetration testing framework, to the China-linked actor it calls Vapor Panda. The campaign targets over 200 high-value organizations worldwide. Source

Anthropic’s VS Code extension on Cursor was backdoored for nearly 12 hours

Attackers compromised Anthropic’s verified Cursor account and pushed a backdoored version of its VS Code extension for close to 12 hours. How the account was accessed is still unknown. Source

Cloud

Cloudflare patched eMudhra weak keys in hours but much of the Internet is still exposed

After certificate authority eMudhra was found on October 19 to have eight trivially weak intermediate keys, anyone could forge certificates for any domain that unpatched clients not enforcing Certificate Transparency would accept. Cloudflare patched its high-impact systems within hours and has logged roughly 22,000 attempted handshakes with the weak keys since October 21, including forged certificates created within hours of the disclosure. It has seen no evidence so far that its own connections were intercepted. Modern browsers are mostly patched, but Android WebView lags with 30-40% still accepting, and more than half of probes on large networks still trust the compromised chain. Patching remains essential. Source

AI

Google Cloud ships Agent Compute Engine, llm-d 1.0 and TPU updates

Google Cloud launched Agent Compute Engine, a compute service that runs autonomous AI agents in hardware-isolated microVM sandboxes with sub-second activation, plus gateway and registry tools for governing them. It also open-sourced llm-d 1.0, a Kubernetes-native stack for serving large language models across TPU and GPU fleets, and TorchTPU with Meta, which brings Google TPUs into PyTorch. GKE gained a preview CPU startup boost with up to 2x faster cold starts, and Oracle will become the first hyperscaler to offer TPUs outside Google Cloud, from 2027. Source

Open source & development

pgAdmin 4 9.19 fixes 15 security vulnerabilities

The release patches 15 CVEs, including an OS command injection in the Import/Export Data tool, LDAP and Kerberos logins landing in the wrong account, and several bypasses of login limits and file-path checks. Server-mode users must configure the new password-command allowlist (SERVER_PASSEXEC_COMMANDS) before first start after upgrading. The release also adds editing data through auto-updatable views and saving query results as JSON or XML. Source

Bottom line

• Install updated Linux kernel packages for CVE-2026-98432 as soon as your distribution ships them. Until then, restrict access to trusted networks or block IPv4 TCP packets with IP options.
• If you run Palo Alto firewalls, apply the PAN-OS PHP2Shell patches and make sure the management interface is not internet-facing.
• Upgrade pgAdmin 4 to 9.19. Server-mode deployments must configure the new password-exec allowlist before first start.