RCE Bugs Pile Up as Attackers Exploit New Flaws

Cloudflare said it detected "millions of exploitation attempts" after the public disclosure of two PHP bugs — a striking metric that captures how fast attackers swarm newly revealed remote-code-execution flaws.

PHP Foundation: PHP2Shell (CVE-2026-109618, CVE-2026-109619)

The PHP Foundation issued patches for two critical vulnerabilities codenamed PHP2Shell — CVE-2026-109618 and CVE-2026-109619 — that can lead to unauthenticated remote code execution. Since the public disclosure, scanning activity has surged: Cloudflare reported millions of attempts, and several websites went offline as a precautionary measure. To date, the source reported no confirmed successful exploitations, but the combination of severity and PHP's broad deployment created an urgent patching imperative.

Supply-chain fallout: LiteLLM leaked master key and MALFEX

Two supply-chain incidents this week underlined how a single secret or a small file can cascade into wide compromise. LiteLLM said an Agent Platform master key accidentally published to a public GitHub repository in late May 2026 enabled an attacker to push a malicious Python package (LiteLLM version 1.107.0). Those files are engineered to run automatically when Python starts and include two payload paths: a credential-harvesting module targeting developer tools, AI apps, cloud environments, and browsers; and a Unix-oriented backdoor that runs as a background process, fetching and executing attacker-supplied code hourly.

Separately, Forever Security described a supply-chain-style campaign codenamed MALFEX in which an AGENTS.md markdown file in a user's Downloads folder tricked an AI coding agent into installing a rogue npm package called "function-color." Forever Security warned that an AGENTS.md is "a file a coding agent is built to trust and obey" and observed that ~/Downloads is a folder a website can deposit a file into "with no click and no permission," letting the web plant instructions an agent will carry out — including rewriting its configuration, installing malware (remote code execution), or handing over login tokens (account takeover).

Node2Shell / Brotli and a parade of critical RCEs

Google's Brotli decoder library received a severe fix for CVE-2026-111497 (also called Node2Shell or HTTP2Shell) that could allow unauthenticated remote code execution. The vulnerability was patched across multiple products and libraries that rely on Brotli, including libbrotli, Python, C++, Node.js, Bun, Envoy, and Istio. Ruby also pushed updates (4.0.9, 3.4.12, and 3.3.13) to address a set of Onigmo regular-expression engine flaws that could lead to remote code execution on GitLab servers (CVE-2026-89078, CVE-2026-93577, CVE-2026-97229, CVE-2026-110264).

At the same time, Plex Media Server vulnerabilities are being chained in the wild as "Plex2Shell" to deliver cryptocurrency miners and proxy software; the U.S. Cybersecurity and Infrastructure Security Agency added all four Plex CVEs to its Known Exploited Vulnerabilities catalog. Linux KSMBD servers were publicly flagged with two CVSS 9.0 remote-code-execution flaws (ZDI-CAN-33774 and ZDI-CAN-33549) and a local privilege-escalation issue (ZDI-CAN-33367); Sysdig urged immediate mitigation such as restricting unprivileged user namespaces and enforcing RuntimeDefault seccomp where possible.

Scanning speed and active exploitation: Redis, MeshCentral, and more

Rapid7's quarterly reporting underscored the shrinking window between disclosure and exploitation: the average time from public high-severity disclosure to observed exploitation is now three days, with 80% of tracked exploitation occurring within 14 days. The week provided concrete examples. Defused Cyber reported observed attempts to exploit CVE-2026-25243 in Redis beginning around October 16, where attackers shaped memory then triggered an over-read and used FLUSHALL to destroy data. The MeshCentral developer warned of an active exploit taking control of servers and remote clients. And CrowdStrike reported multiple previously unreported operators using "Claude Code" agentic AI to plan and execute campaigns across healthcare, government, and private sectors.

What this means for technologists, enterprise ops, and end users

• Technologists and security teams: Patch first — PHP2Shell, Brotli (CVE-2026-111497), GitLab CVE-2026-110264 (patched in multiple 19.x and 18.11.13 releases), and any listed high-risk CVEs. Follow Sysdig’s guidance for KSMBD: restrict unprivileged namespaces and apply RuntimeDefault seccomp where possible.
• Developers and package maintainers: Treat leaked keys as urgent — LiteLLM's case shows a single exposed master key can permit backdoored releases that run on Python startup. Also note ecosystem responses: Python maintainers removed an affected root CA from cert bundles and eMudhra exited certificate issuance after weak keys were blocked by major browsers.
• End users and product owners: Revoke and rotate exposed credentials — Railway urged users to revoke sensitive environment variables stored before October 21, 2026, after a campaign aimed at obtaining AI provider API keys. Be cautious with AI tools and downloaded files: Johns Hopkins Bayview reported patient data exposure after a staff member uploaded a spreadsheet to an unapproved AI platform; Forever Security showed how a markdown file in Downloads can weaponize an AI coding agent.

The practical lesson in the week's reporting is concrete and uncomfortable: unauthenticated remote-code-execution bugs are numerous, scanning and exploitation are fast, and simple operational mistakes — a leaked key, an unvetted markdown file, an unpatched runtime — can hand attackers an easy path. With millions of probes detected and average exploitation measured in days, the sensible posture is immediate patching, credential rotation, and treating trusted defaults as untrusted until verified.

Original story at The Hacker News