A roundup of 255 items curated from across the security community.

PHP2Shell: Critical Pre-Auth RCE in PHP (CVE-2026-109618)

Calif.io discovered two critical PHP vulnerabilities enabling unauthenticated remote code execution on stock configurations. Trail of Bits confirmed a production-grade exploit hitting WordPress, MediaWiki, phpBB, and Laravel across both ARM64 and x86-64. Unit 42 confirmed active exploitation in the wild. NSA, FBI, and CISA all recommended taking affected systems offline.

Kuba Gretzky · Read more

+4 more on this storyPHP 8.5.12 released - 90 fixes in 29 components, including security fixes (CVE-2026-109618, CVE-2026-109619) (Benjamin Delpy)
Amazon says PHP2Shell exploitation began 4 days before patches (BleepingComputer)
‼ ok, this might be the first publicly documented, validated PHP2Shell RCE in the wild (Dan Guido)
If Andy says so, i would take this from granted. No need for details. (Giuseppe `N3mes1s`)

News

Private keys for eight TLS certificates were trivially small integers (4, 56, 66, 115, 155, 156, 230, 237). An intermediate CA compromise affecting the web PKI trust chain.

+2 more on this storyCompromised eMudhra intermediate certificates (Cisco Talos)
I used slicervm as a forensic environment to explore what MITM would look like with the compromised eMudhra... (Giuseppe `N3mes1s`)

Versions v0.11.0 through v0.15.0 of the OVHcloud CLI were compromised by a hacker. Detected and blocked within 90 minutes. No other repos affected.

Reuters reports a single attacker unleashed AI agents on 1,200+ websites, compromising over 100 for roughly $50 per hack in token costs. IR firms are saturated and turning away work.

Microsoft attributes the ConnectWise supply chain compromise to Chinese state-aligned Storm-0940. The group used the agentic AI pentesting framework ARTEX to automate attacks across hundreds of organizations.

+1 more on this storyThrilled to share my latest research on AI-assisted cyberattacks and the ARTEX hacking framework! (Giuseppe `N3mes1s`)

CVE-2026-110264 allows unauthenticated remote code execution on self-managed GitLab servers. The regex compiler flaw spans every release in the past 12 years.

+1 more on this storyRegexp compilation fixes related to GitLab's CVE-2026-89078, CVE-2026-93577, CVE-2026-97229 and CVE-2026-110264 (Nicolas Krassas)

Politico reveals TSO, a new DoD agency with significant budget and appetite for offensive hacking tools, spyware, and sabotage capabilities.

Multiple vulnerabilities in BIND 9, the most widely deployed DNS software on the internet, including remote DoS and memory corruption.

A second critical PHP RCE dropped just two days after the emergency PHP2Shell patches, compounding the patching emergency for PHP operators.

Eye Security found a Popcorn Time updater connecting to an abandoned domain. Anyone who registered it could push arbitrary executables to 3.8 million clients with elevated privileges.

More this week (50)

Techniques and Write-ups

SpecterOps found that LDAP and SMB signing can be bypassed through re-authentication on attacker-owned connections. Signing still enforces, just with the attacker’s session key. Microsoft patched SMB silently in September and LDAP in October.

Datadog Security Labs found a technique that tanks Jev’s prompt-injection detection rate and bypasses LangChain’s experimental tool-risk gating.

Horizon3 deep dive on a Fortinet FortiSIEM vulnerability chaining gRPC path traversal with arbitrary file write to achieve remote code execution.

+1 more on this storyProof of concept exploit for CVE-2026-106173, a path traversal file write leading to RCE for Fortinet FortiSIEM (Nicolas Krassas)

SpecterOps researchers harmj0y and tifkin_ demonstrate how adversarial machine learning can evade EDR detection, and how LLMs accelerate the reverse engineering behind it.

FoxIO discovered a QPACK amplification attack that lets anyone remotely crash servers running vulnerable HTTP/3 libraries with a single request.

CVE-2026-111497. A single Brotli-compressed request to any Express app using express.json() achieves code execution. Affects Node.js 22, 24, and 26.

+3 more on this storyWe’re still testing how exploitable the Node.js/Brotli bug is, but we’ve found that it’s surprisingly hard to tell... (Giuseppe `N3mes1s`)
The Broccoli bug (CVE-2026-111497) affected Chrome for ~8 years (Halvar Flake)
Node2Shell PoC is now public (kmkz)

Independent research abusing documented SMB re-authentication behavior to defeat enforced signing. PoC released. Patched on recent Windows versions.

Unprompted AU talk walking through practical AI-assisted exploit development against XNU, from vulnerability analysis to working exploit.

Working exploit under stock php.ini-production settings with ASLR, multi-worker, and load. Targets WordPress, MediaWiki, phpBB, and Laravel on Apache and nginx, both architectures.

Captures Windows Hello for Business PINs by placing a CPU hardware breakpoint on LogonUI.exe’s PIN packaging function. Login succeeds normally; the PIN is read from the trapped thread context.

More this week (115)

Tools and Exploits

Tool to check if your WiFi device is vulnerable to the unpatched Broadcom wormable RCE. Known affected devices listed.

SLCyber releases a high-speed HTTP/3 fuzzer capable of 100K requests per second for finding QUIC and HTTP/3 implementation bugs.

Ludus lab role to reproduce the eMudhra CA compromise. Mint EV certs for any domain and code-sign any binary in your lab.

Datadog Security Labs released a framework for discovering and exploiting AWS IAM privilege escalation paths.

Full source release of an autonomous root chain for TCL Android TV exploiting CVE-2022-22706 in the Mali GPU driver.

Next generation of the SysWhispers direct syscall tool for evading EDR userland hooks.

Assetnote’s high-fidelity detection scanner for CVE-2026-111497 (Node2Shell).

Writeup on escaping the Python sandbox in AWS’s MCP server implementation.

The largest update in hashcat history. PCFG attacks let you crack passwords using grammar-based models, hash recipes eliminate the need for fixed hash-modes, and bcrypt now runs on FPGA boards.

Tracebit introduces context bombs as a defensive technique against agentic AI attackers, poisoning their context windows to burn tokens and derail autonomous exploitation.

More this week (35)