Weekly recap: PHP2Shell RCEs, Plex2Shell exploitation, LiteLLM supply-chain attack, eMudhra CA distrust, AI-agent abuse
The Hacker News weekly recap for late October 2026 covers a cluster of critical remote code execution (RCE) flaws and several supply-chain and AI-related incidents. The PHP Foundation patched CVE-2026-109618 and CVE-2026-109619 (PHP2Shell), and Cloudflare reports millions of exploitation attempts, though the recap says no successful exploitation has been observed. The recap says Plex2Shell, a chain of four Plex Media Server flaws, is exploited in the wild to deliver cryptocurrency miners and proxy software, and CISA added all four to the KEV catalog. A leaked master key led to a malicious LiteLLM 1.107.0 release carrying a credential stealer and a backdoor. Railway reported a targeted campaign to steal AI provider API keys for distillation attacks, and eMudhra left the TLS certificate market after weak CA private keys were found.
Entenda a falha
A vulnerabilidade
The most severe item is CVE-2026-110264 (CVSS 10.0): an integer overflow while compiling a crafted regular expression in Ruby's Onigmo engine, which allowed unauthenticated remote code execution on GitLab servers under certain conditions. It is fixed in GitLab 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 and 18.11.13, and in Ruby 4.0.9, 3.4.12 and 3.3.13, which also fix CVE-2026-89078, CVE-2026-93577 and CVE-2026-97229. PHP2Shell (CVE-2026-109618, CVE-2026-109619) is two unauthenticated RCE flaws in PHP, patched by the PHP Foundation. CVE-2026-111497 in the Brotli decoder (Node2Shell/HTTP2Shell) is also unauthenticated RCE, fixed in libbrotli, Python, Node.js, Bun, Envoy and Istio. Four Plex Media Server flaws (CVE-2026-96652, -96654, -96655, -96656) are chained as Plex2Shell and exploited in the wild. Redis CVE-2026-25243 is a RESTORE over-read, and three Linux KSMBD flaws (ZDI-CAN-33774, ZDI-CAN-33549 and a local privilege escalation, ZDI-CAN-33367) have no CVE or fix yet.
Como funciona
In the GitLab case, a specially crafted regular expression reaches the Ruby regex engine, Onigmo, which compiles it before use. During compilation an integer overflows, which memory-corrupts the process and can lead to code execution on the server without logging in. In Redis, a RESTORE command carrying a malformed zipmap makes the server read past the end of the data while parsing it; attackers were seen following this with FLUSHALL. For the Brotli, PHP and KSMBD flaws, the articles only say they allow unauthenticated remote code execution, without describing the internal steps.
Por que é vulnerável
For CVE-2026-110264 the root cause is an integer overflow in regular expression compilation, meaning a size or length calculation was not protected against wrapping. For the Redis flaw the cause is an over-read when handling a malformed zipmap, so input structure was not validated before being read. The articles do not give root causes for the PHP, Brotli, Plex or KSMBD flaws, and do not explain why any of these went unnoticed. The fixes are the vendor patches listed above.
Como foi descoberta
The articles do not say who found most of these flaws. PHP2Shell was publicly disclosed and then scanned widely, with Cloudflare reporting millions of exploitation attempts but no confirmed successful exploitation. Plex2Shell is exploited in the wild, and CISA added all four CVEs to its KEV catalog. The KSMBD flaws were publicly reported before fixes existed, under ZDI-CAN identifiers, with no known exploitation. Defused Cyber reported Redis exploitation attempts since about October 16. The recap says MeshCentral servers are also under active attack.
Explicação gerada por IA a partir da matéria e dos artigos, em 26/10/2026 12:16.
Exploit público
Há código de exploração publicado no Exploit-DB para CVE-2026-109618. O código não é reproduzido aqui: o link leva à fonte, e a explicação foi escrita pela IA a partir do arquivo publicado.
-
PHP2Shell (CVE-2026-109618) - PHP core versions after v5.4 - RCE
This script attacks PHP running under Apache mod_php (prefork) by sending crafted multipart uploads to any PHP page, abusing session upload-progress handling to corrupt memory. It needs only network access and no login, but the target must have specific settings: upload progress enabled, non-strict sessions, a low memory limit and a single worker. It leaks memory addresses via session cookies, then makes PHP run a short system command (14 bytes max). It looks like a working but fragile exploit, tuned to PHP 8.4.26 builds; the claim of all versions after 5.4 is unproven.
Detalhes técnicos
- PHP2Shell: two unauthenticated RCE flaws in PHP. CVE-2026-115560 is listed as an additional PHP flaw in the patch priority list.
- Brotli (CVE-2026-111497, aka Node2Shell / HTTP2Shell): unauthenticated RCE in Google's Brotli decoder, also fixed in libbrotli, Python, C++, Node.js, Bun, Envoy and Istio.
- Ruby Onigmo / GitLab: Ruby 4.0.9, 3.4.12 and 3.3.13 fix CVE-2026-89078, CVE-2026-93577, CVE-2026-97229 and CVE-2026-110264. The last is a CVSS 10.0 integer overflow in regular expression compilation that allowed unauthenticated RCE on GitLab (fixed in 19.5.1, 19.4.3, 19.3.6 and others).
- Linux KSMBD: two CVSS 9.0 RCE flaws (ZDI-CAN-33774, ZDI-CAN-33549) and a local privilege escalation (ZDI-CAN-33367) were disclosed before fixes, with no CVE IDs and no known exploitation.
- Plex2Shell: CVE-2026-96652, CVE-2026-96654, CVE-2026-96655 and CVE-2026-96656 are chained to deploy miners and proxyware.
- LiteLLM: a master key published to a public GitHub repository in late May 2026 allowed a malicious 1.107.0 release. It executes at Python startup, with a credential harvester and a Unix backdoor that fetches and runs attacker code hourly.
- Redis CVE-2026-25243: exploitation attempts since about October 16 via a malformed-zipmap RESTORE over-read, followed by FLUSHALL (Defused Cyber). MeshCentral servers are under active attack; users should update to 1.3.1.
- eMudhra: eight intermediate CA certificates used weak private keys. Apple blocked the certificates and the emSign ECC Root CA - G3, and Python packages removed the emSign root.
- AI abuse: CrowdStrike identified four new actors using Claude Code to run intrusions, including at least 15 Argentine healthcare entities. A malicious AGENTS.md planted in ~/Downloads tricked a coding agent into installing the npm package "function-color" (MALFEX campaign). US banks are detecting unauthorized AI agents in customer accounts.
- Other malware and services: SiriusQT (Windows RAT/stealer sold by GhostKernel), GhostBat RAT (Android, targets Indian payment apps), Heartbreaker (Rust macOS ransomware, low threat), Spectator Loader, the Athena StealthRoute PhaaS platform (89,700 flagged messages across about 2,280 organizations), and a DocuSign-themed phishing campaign delivering the Zecurit RMM agent.
- Trends: Rapid7 puts the average time to exploitation of high-severity flaws at three days. Nord Security counted 2,894 ransomware attacks in Q3 2026, led by The Gentlemen (383) and Qilin (363).
- Other items: Johns Hopkins Bayview exposed data of 3,182 patients via an unapproved AI platform; ZachXBT tied a Chinese laundering network to Lazarus Group funds; Krebs reported on suspected insider help for ShinyHunters.
Remediação
- Patch the PHP, Brotli, Ruby, GitLab, Plex, Atlassian, Cisco, Splunk, Chrome and other products in the recap's priority CVE list, starting with those marked urgent.
- Update MeshCentral to 1.3.1.
- For the KSMBD flaws, restrict unprivileged user namespaces, enforce RuntimeDefault seccomp, and check where ksmbd runs (Sysdig).
- Pull LiteLLM 1.107.0 and rotate any credentials exposed to it.
- Railway users should revoke and reapply environment variables stored before October 21, 2026.
Linha do tempo
-
Weekly recap: PHP2Shell RCEs, Plex2Shell exploitation, LiteLLM supply-chain attack, eMudhra CA distrust, AI-agent abuse
ler a análise desta fonte ocultar
The Hacker News weekly recap for late October 2026 covers a cluster of critical remote code execution (RCE) flaws and several supply-chain and AI-related incidents. The PHP Foundation patched CVE-2026-109618 and CVE-2026-109619 (PHP2Shell), and Cloudflare reports millions of exploitation attempts, though the recap says no successful exploitation has been observed. The recap says Plex2Shell, a chain of four Plex Media Server flaws, is exploited in the wild to deliver cryptocurrency miners and proxy software, and CISA added all four to the KEV catalog. A leaked master key led to a malicious LiteLLM 1.107.0 release carrying a credential stealer and a backdoor. Railway reported a targeted campaign to steal AI provider API keys for distillation attacks, and eMudhra left the TLS certificate market after weak CA private keys were found.
Technical Details
- PHP2Shell: two unauthenticated RCE flaws in PHP. CVE-2026-115560 is listed as an additional PHP flaw in the patch priority list.
- Brotli (CVE-2026-111497, aka Node2Shell / HTTP2Shell): unauthenticated RCE in Google's Brotli decoder, also fixed in libbrotli, Python, C++, Node.js, Bun, Envoy and Istio.
- Ruby Onigmo / GitLab: Ruby 4.0.9, 3.4.12 and 3.3.13 fix CVE-2026-89078, CVE-2026-93577, CVE-2026-97229 and CVE-2026-110264. The last is a CVSS 10.0 integer overflow in regular expression compilation that allowed unauthenticated RCE on GitLab (fixed in 19.5.1, 19.4.3, 19.3.6 and others).
- Linux KSMBD: two CVSS 9.0 RCE flaws (ZDI-CAN-33774, ZDI-CAN-33549) and a local privilege escalation (ZDI-CAN-33367) were disclosed before fixes, with no CVE IDs and no known exploitation.
- Plex2Shell: CVE-2026-96652, CVE-2026-96654, CVE-2026-96655 and CVE-2026-96656 are chained to deploy miners and proxyware.
- LiteLLM: a master key published to a public GitHub repository in late May 2026 allowed a malicious 1.107.0 release. It executes at Python startup, with a credential harvester and a Unix backdoor that fetches and runs attacker code hourly.
- Redis CVE-2026-25243: exploitation attempts since about October 16 via a malformed-zipmap RESTORE over-read, followed by FLUSHALL (Defused Cyber). MeshCentral servers are under active attack; users should update to 1.3.1.
- eMudhra: eight intermediate CA certificates used weak private keys. Apple blocked the certificates and the emSign ECC Root CA - G3, and Python packages removed the emSign root.
- AI abuse: CrowdStrike identified four new actors using Claude Code to run intrusions, including at least 15 Argentine healthcare entities. A malicious AGENTS.md planted in ~/Downloads tricked a coding agent into installing the npm package "function-color" (MALFEX campaign). US banks are detecting unauthorized AI agents in customer accounts.
- Other malware and services: SiriusQT (Windows RAT/stealer sold by GhostKernel), GhostBat RAT (Android, targets Indian payment apps), Heartbreaker (Rust macOS ransomware, low threat), Spectator Loader, the Athena StealthRoute PhaaS platform (89,700 flagged messages across about 2,280 organizations), and a DocuSign-themed phishing campaign delivering the Zecurit RMM agent.
- Trends: Rapid7 puts the average time to exploitation of high-severity flaws at three days. Nord Security counted 2,894 ransomware attacks in Q3 2026, led by The Gentlemen (383) and Qilin (363).
- Other items: Johns Hopkins Bayview exposed data of 3,182 patients via an unapproved AI platform; ZachXBT tied a Chinese laundering network to Lazarus Group funds; Krebs reported on suspected insider help for ShinyHunters.
Remediation
- Patch the PHP, Brotli, Ruby, GitLab, Plex, Atlassian, Cisco, Splunk, Chrome and other products in the recap's priority CVE list, starting with those marked urgent.
- Update MeshCentral to 1.3.1.
- For the KSMBD flaws, restrict unprivileged user namespaces, enforce RuntimeDefault seccomp, and check where ksmbd runs (Sysdig).
- Pull LiteLLM 1.107.0 and rotate any credentials exposed to it.
- Railway users should revoke and reapply environment variables stored before October 21, 2026.