Goldberg Security Research is an independent IT and security consultancy — penetration testing, vulnerability research, and responsible disclosure. Our published work speaks for itself.
Who we are.
Goldberg Security Research provides information technology consultancy and services with a security-first focus. We work across vulnerability research, penetration testing, reverse engineering, and low-level systems security — and we publish our findings. We also develop our own analysis tooling, including a custom Type-1 VT-x hypervisor purpose-built for examining malware that actively evades commercial sandboxes and off-the-shelf virtualisation. Everything here is conducted against our own infrastructure or under authorization, and disclosed responsibly.
IT consultancy & security services.
Penetration testing and security reviews of applications, networks, and infrastructure — with clear, prioritised, actionable reporting.
Deep-dive research, reverse engineering, and responsible disclosure. We analyse malware first-hand in our own lab using a custom VT-x research hypervisor built to defeat anti-analysis evasion.
Advisory on secure architecture, secure development, and technology decisions — tailored to your environment and risk profile.
Custom security tooling — from kernel drivers and analysis hypervisors to automation pipelines — and broader IT services to support your team's day-to-day needs.
The people behind the work.

Leads Goldberg Security Research. Specialises in malware analysis, reverse engineering and low-level systems research — including the development of our in-house VT-x research hypervisor. Has discovered and disclosed vulnerabilities in Windows kernel components and npm packages.
Generalist security researcher working across web, application, and infrastructure security — wherever the interesting bugs are.
Broad-spectrum researcher spanning reverse engineering, vulnerability discovery, and the tooling that supports them.
All-rounder across offensive security, exploit research, and automation — comfortable from kernel to cloud.
Original malware research and independent re-analysis of notable public vulnerabilities — root cause, exploitation, and defence. (First-hand analyses are labelled as such; re-analyses credit the original researchers.)
The first self-propagating npm worm: stolen maintainer tokens auto-republish infected packages while a postinstall payload harvests secrets to public GitHub repos. 500+ packages hit.
A prolific MaaS stealer using "ClickFix" lures and Telegram/Steam dead-drop resolvers for resilient C2 — and how it survived the May 2025 Microsoft/DOJ takedown.
First-hand analysis of Lumma samples sourced from YouTube "cheat" videos — Fortnite aimbots, Roblox executors, CoD unlock tools — examined in our hypervisor with Ghidra and Wireshark. Includes anti-VM observations.
A fake "Donut SMP Dupe" Fabric mod that steals Minecraft credentials on load, then fetches a JNIC-compiled Stage 2 entirely in-memory — with C2 resolved via an RSA-signed Ethereum smart contract.
The open-source Go C2 adopted by APT29 and ransomware crews as a Cobalt Strike alternative — unique per-build implants, multi-protocol C2, and where memory analysis beats it.
The industrialised RaaS brand — intermittent encryption for speed, DLL-reflection evasion, and the Operation Cronos takedown that seized its source code and decryption keys.
A 15-year banking-trojan-turned-loader and primary ransomware on-ramp — hijacked email threads, OneNote/PDF lures. Taken down in 2023 (700k+ devices), back within months.
One of the most prolific .NET stealers — browser creds, wallets, screenshots, and Lua-bytecode evasion. Infrastructure seized in Operation Magnus (2024); variants persist.
The commodity RAT behind countless forks — keylogging, screen/audio capture, multi-stage script delivery, and living-off-the-land persistence.
A USB-spread loader with elite anti-analysis — ETW blinding, hook detection, opaque predicates — that sells footholds leading to Dridex and ransomware.
ndis.sys Use-After-Free LPE
A use-after-free in the Windows NDIS driver lets a local attacker escalate to kernel code execution. Reported via ZDI, patched by Microsoft March 2026. Discovered by Goldberg Security Research.
Async functions return a globalPromise whose .catch callbacks are not
sanitised — a Symbol-named Error triggers a host TypeError on that path, yielding the host Function
constructor and full RCE. Discovered by Goldberg Security Research.
A Symbol.species hijack on a host-realm Promise plus V8's internal
Thrower substitution bypasses vm2's rejection sanitizer entirely — delivering a raw host
object to sandbox code for full RCE. Discovered by Goldberg Security Research.
A multi-year social-engineering campaign planted a hidden backdoor in a core compression library to compromise OpenSSH pre-auth — caught days before mass rollout by a 500ms slowdown.
One logged string — ${jndi:ldap://…} — turned a ubiquitous Java logging
library into unauthenticated RCE across hundreds of millions of systems.
A reintroduced signal-handler race in sshd gives unauthenticated root RCE in
the default config — hard to win, but ~14M instances were exposed at disclosure.
A double-free in the netfilter nf_tables verdict path lets an
unprivileged local user corrupt kernel memory and escalate to root. Actively exploited in the wild.
An uninitialised pipe-buffer flag lets an unprivileged process write into the page
cache of read-only files — overwrite a setuid binary or /etc/passwd and you have root.
below Monitor
A world-writable log directory plus a symlink-following log writer lets a local
user redirect privileged writes into /etc/passwd and escalate to root. A clean logic bug.
A frame-count vs. parse-time mismatch in the USB-camera driver lets a malicious USB device write past a kernel heap buffer. Exploited in the wild (CISA KEV).
Catastrophic regex backtracking in one of npm's most-depended-on packages — a crafted string pins a CPU core and hangs the process. Deep in countless dependency trees.
A buffer overflow in ld.so's GLIBC_TUNABLES parsing gives
local root via any setuid binary on default Fedora/Ubuntu/Debian installs.
An || where an && belonged sent the anti-CSRF token to
every host when withCredentials was set — handing it to third parties.
A {} index instead of Object.create(null) lets a
__proto__ cookie pollute the global prototype across the whole app.
Engagements, consultancy, or responsible disclosure.