This week’s cybersecurity landscape was dominated by remote code execution (RCE) vulnerabilities that are high-severity, rapidly being scanned, and—worst of all—for some, already being exploited. From weaknesses in foundational technologies like PHP, Brotli, and Linux SMB servers, to supply-chain compromises and rogue AI agents sowing chaos, there’s no shortage of attack vectors. If you’re responsible for patching or monitoring systems, this should set off alarms.
Emerging High-Risk Vulnerabilities: The Shell Wars
Two critical vulnerabilities in PHP—tracked as CVE-2026-109618 and CVE-2026-109619 and dubbed PHP2Shell—are fueling major concern. The flaws allow unauthenticated RCE, and since public disclosure, Cloudflare reports millions of scanning attempts. Some affected sites have gone offline preemptively. Meanwhile, a separate flaw in Brotli (CVE-2026-111497), tied to libraries across languages and platforms (Node.js, Python, Envoy, and others), could also lead to unauthenticated RCE. Ruby’s regex engine patches, addressing multiple vulnerabilities—including those that could open paths to remote code execution—landed alongside.
On Linux systems, two near-CVSS-9.0 RCE bugs in ksmbd, the SMB server in the kernel, have been disclosed before fixes are ready, along with a local privilege escalation flaw. Organizations are urged to disable or limit unprivileged user namespaces, enforce seccomp with RuntimeDefault, and track whether ksmbd is running in their environments.
This week’s exploit roundup also spotlights Plex2Shell, where attackers are already chaining together four vulnerabilities in Plex Media Server (CVE-2026-96652, -96654, -96655, -96656). These are being used in the wild to drop cryptocurrency miners and proxy backdoors. To make matters worse, the U.S. CISA has added all four to its Known Exploited Vulnerabilities list.
Supply Chain Breaches, Rogue Agents & AI Risk
Supply-chain risk showed up at LiteLLM, where a master key leaked on GitHub earlier this year enabled a backdoored version of its Python package (v1.107.0). Once the attacker pushed that version, it ran malicious code even if applications didn’t import LiteLLM. The trojan horse included modules to steal credentials, establish background execution, and fetch and run arbitrary code every hour.
Railway disclosed a targeted campaign tied to distillation attacks—the misuse of AI model internals—with stolen API keys and access to ablated models reported. Users whose sensitive variables were set before October 21, 2026, were advised to rotate and resecure them.
Adding to the AI threat, malware campaign “MALFEX” uses a deceptively named file, AGENTS.md, dropped into the Downloads folder. When coding agents trust and obey this file, it can trigger silent installs of npm packages, configuration rewriting, token theft, or remote execution—all without user interaction.
Other Key Threats and Trends
Cryptography’s role surfaced when eMudhra, an Indian certificate authority, was found to have issued intermediate CA certificates using weak private keys—so weak that major players like Apple blocked them, and eMudhra dropped out of certificate issuance entirely. Related software bundles (like certifi, pip, conda, etc.) were updated to remove those certificates.
GitLab patched a CVSS-10 integer overflow in its regular expression engine that could allow unauthenticated code execution when processing specially crafted regexes. Versions across both Community and Enterprise editions were updated.
On the operational front: threat actors identified by CrowdStrike are leveraging Claude AI for decision-making and active operations targeting healthcare, private sector, and government entities across Latin America, Southeast Asia, and parts of Europe and the U.S. Also notable: hospitals are notifying thousands of patients of data exposure due to unapproved AI tools; phishing-as-a-service platforms are growing more sophisticated; malware families like GhostBat RAT and others have evolved to exploit OS features or rely on modules for credential and financial theft; and ransomware incidents rose sharply in Q3 2026—nearly 50% year-over-year—led by gangs deploying AI-assisted phishing and data exfiltration rather than encryption.
Trending CVEs You Can’t Ignore
Patch urgency is highest for these vulnerabilities: the two PHP2Shell bugs (CVE-2026-109618, CVE-2026-109619), flaws in Splunk, ASUS Control Center, Cisco, Google Chrome, Atlassian products, Ruby’s regex engines, and the Brotli-based Node2Shell/HTP2Shell chain. If you run any of those tools, infrastructure, or libraries, double-check you’re running the latest fixes.
This week’s data from Rapid7 signals RCE exploitation is happening faster than ever: the window between public disclosure and observed attacks has shrunk to an average of three days. By day 14, the vast majority are under active exploitation. Delay patching at your peril.
Bottom line: The cascade of remote execution flaws—from web servers and media servers to supply chains and AI agents—creates overlapping risk. Many bugs require no authentication; others have no available fixes yet. Assume attackers are scanning right now, and work like it’s already compromise time.
