AtlasLatino — research feed

• August 9, 2026 Arsenal Offensive Security AppSec Village — DEF CON 2026 Lenin Alevski MX ROP for the Web: Smuggling XSS, SQLi, and Web Shells Past Every WAF Using Compression Dictionaries

  Compression Dictionary Transport (RFC 9842) shipped in Chrome 130+ in 2025. Any JavaScript asset designated as a dictionary becomes a set of gadgets an attacker can chain into responses that bypass signature-based WAFs and IDSes. gadget-scanner is the a tool to evaluate that surface: drop in a candidate dictionary and a payload, and it reports how much of the payload hides inside backreferences and how much leaks as literal bytes on the wire. At the table you will score real-world JS libraries as attack primitives, generate the DCB blob with brotli -D, and watch the side-by-side: the raw bytes a WAF would see vs. the payload a browser would execute

  ResearchersLenin Alevski MX

  Presented at AppSec Village — DEF CON 2026

  View official session
• August 9, 2026 Talk Offensive Security Red Team Village — DEF CON 2026 Murilo Caixeta BR Living Off WebView2: Turning Microsoft's Browser into a Red Team Asset

  Modern Windows applications are quietly shipping with a built-in Chromium-based browser, WebView2, running inside trusted, signed processes such as Microsoft Teams, Outlook, and other enterprise software.

  This talk shows how that design choice creates a powerful and largely unmonitored attack surface.

  I demonstrate how this model can be abused in offensive operations, turning WebView2 into a vector for persistence, code execution, and enterprise account impersonation.

  I show how feature flags and environment variables can be leveraged to manipulate the WebView2 runtime, enabling techniques such as DLL sideloading within legitimate applications. This allows arbitrary code execution inside trusted processes, supporting stealthy living-off-the-land persistence.

  Building on this, I present methods to intercept and proxy HTTP/HTTPS traffic generated by WebView2-based applications. This enables the extraction of session tokens, cookies, and other sensitive artifacts, leading to realistic account takeover and impersonation scenarios, including access to platforms such as Office 365.

  In addition to the offensive techniques, I provide a detailed analysis of the Indicators of Compromise (IOCs) generated throughout these attack chains. I highlight detection opportunities, discuss current visibility gaps in EDR solutions, and propose practical approaches for identifying and responding to this type of activity in real-world environments.

  ResearchersMurilo Caixeta BR

  Presented at Red Team Village — DEF CON 2026

  View official session
• August 9, 2026 Talk Malware & Reversing Red Team Village — DEF CON 2026 Eduardo Chavarro Ovalle CO·Ashley Hiram MX·Alexander Rodchenko Evil Is Always a Bad Stylist: .NET Obfuscation with Roslyn

  Modern endpoint defenses increasingly fingerprint .NET tools not only by strings and imports, but by structure: control flow, call graphs, and overall execution "workflows." Traditional IL obfuscation can hide obvious indicators, yet often produces highly regular flattened patterns that are easy to flag on their own.

  This talk presents a Roslyn-based, source-to-source obfuscator for C# that rewrites an entire project before compilation. Using the official compiler APIs, it transforms the code's syntax and semantic structure, reshapes control/data flow, and then recompiles the result - while keeping the program's observable behavior intact. The output tends to look complex but "natural," diverging both from the original tool and from the telltale patterns created by many IL-level obfuscators.

  I will walk through the pipeline, show how surprisingly little code is needed to build these transformations on Roslyn, and discuss what this means for EDR, sandboxing, and ML detections that rely on graph and workflow analysis of .NET binaries.

  So, anyway, on moment of submitting original Seatbelt tool (obfuscated by source) have 10/73 security vendors flagged this file as malicious (can be better).

  My project https://github.com/gam4er/Loaders differs from more well known https://github.com/sadreck/Codecepticon by implementing semantic structure obfuscation.

  My point of view (as a SOC analyst, blue teamer) on "how to hide my tool from AV" problem is: if you have a code — OBFUSCATE YOUR CODE. Stop modifying binaries.

  ResearchersEduardo Chavarro Ovalle CO·Ashley Hiram MX·Alexander Rodchenko

  Presented at Red Team Village — DEF CON 2026

  View official session Loaders on GitHub
• August 9, 2026 Talk Other Bug Bounty Village — DEF CON 2026 Ariel Walter García AR Heavy Metal and Hidden Secrets: A Five-Year Retrospective on DEF CON Challenge Coins

  Take a look behind the scenes at five years of custom DEF CON challenge coins from DEF CON 30 through DEF CON 34. This session walks through the entire lifecycle of these unique hardware puzzles, exploring the inspiration behind their designs, step-by-step solutions, and how the rise of AI is shifting the future of cryptographic challenges.

  ResearchersAriel Walter García AR

  Presented at Bug Bounty Village — DEF CON 2026

  View official session
• August 9, 2026 Talk Incident Response La Villa Hacker — DEF CON 2026 Victor Gomez PA·Horacio Bazán PA La Rosa de Guadalupe DFIR: Serie de 3 capítulos de misterio Cyber

  Tres insólitos incidentes de ciberseguridad fueron resueltos por dos investigadores en Incidentes de Ciberseguridad con ayuda de múltiples técnicas de "sniper-forensics" y una "Rosa Blanca" les confirmó sus sospechas iniciales y los llevó a encontrar la verdad. Se abordará el análisis detallado de las diferentes fuentes de información (bitácoras, artefactos y contexto).

  ResearchersVictor Gomez PA·Horacio Bazán PA

  Presented at La Villa Hacker — DEF CON 2026

  Speaker page (Victor Gomez) Speaker page (Horacio Bazán)
• August 9, 2026 Talk AI Security La Villa Hacker — DEF CON 2026 Alcyon Junior BR Bait the Bot: Hacking Back Autonomous AI Vulnerability Scanners

  Autonomous, LLM-powered vulnerability scanners now crawl the web at scale, and every one of them shares the same structural weakness: it reads the content you control. This talk flips the engagement. Instead of just blocking the bot, your site detects that an AI agent is scanning it, fingerprints the model behind it, and fires back a hidden, tailored prompt injection that derails, poisons, or neutralizes the attacking agent, with no traditional exploit involved.

  We build on recent academic work on prompt-injection "hack back," push it into the reconnaissance-and-scanning threat model with live demos, walk through an escalating cat-and-mouse between attacker and defender, and take an honest look at where active defense crosses the legal line. Red teamers and blue teamers alike will leave with something they can responsibly test the same week.

  ResearchersAlcyon Junior BR

  Presented at La Villa Hacker — DEF CON 2026

  View official session
• August 9, 2026 Workshop Offensive Security Red Team Village — DEF CON 2026 Fernando Mengali BR·Thiago Cunha da Silva BR From Buffer Overflow to Blackout: Chaining Attacks Against Industrial Systems

  This presentation takes a practical, demo-driven approach to exploitation, demonstrating how a buffer overflow vulnerability can still serve as an entry point to compromise modern industrial infrastructures. The session is approximately 90% hands-on and 10% theoretical.

  The talk walks through the full attack chain, starting from initial exploitation and progressing to the development and execution of multistage shellcode. This shellcode establishes communication with a remote Command and Control (C2) server, retrieves an additional payload, and executes it directly in memory, bypassing traditional detection mechanisms.

  Once the environment is compromised, the session demonstrates how an attacker can gain full remote control, enabling command execution, lateral movement, and exfiltration of sensitive data, including critical assets such as licenses and proprietary industrial information.

  The talk also includes interaction with real-world devices such as RTUs (Remote Terminal Units), showing how malicious actions can directly impact industrial operations. In critical scenarios, this can lead to disruptions in essential services, including failures in energy systems and potential blackouts.

  The content is based on real-world pentesting experience conducted in an energy sector company in Brazil, providing a practical and applied perspective on risks in ICS/SCADA environments.

  The goal is to demonstrate how modern attacks combine vulnerabilities with advanced techniques, reinforcing the need for robust security strategies to protect critical infrastructure.

  ResearchersFernando Mengali BR·Thiago Cunha da Silva BR

  Presented at Red Team Village — DEF CON 2026

  View official session
• August 9, 2026 Talk Offensive Security Red Team Village — DEF CON 2026 Elzer Pineda PA·Jose Manuel Rivas PA Tokens and PRT: Advanced Attacks and Persistence in Microsoft Entra ID

  Is MFA and Conditional Access a real security guarantee? In this technical session, we will demonstrate how endpoint compromise allows an attacker to bypass traditional identity barriers in the cloud. The talk focuses on exploiting vulnerabilities in Microsoft Entra ID, breaking down an advanced attack chain:

  • Device Code Flow: Abuse of authentication flows for initial access (Demo with Entraith).
  • PowerShell Hijacking: Process interception to obtain session tokens (GrabTokenAzureAD).
  • Sliver BOF Extraction: Use of Beacon Object Files (BOF) for stealthy exfiltration of tokens and the Primary Refresh Token (PRT) from memory, evading anti-malware defenses.
  • MFA Bypass and Intune: The speakers developed custom tooling to extract local PRTs, bypass Intune device management controls, and circumvent MFA. This entire attack lifecycle was consolidated into Entraith — an offensive framework built from scratch by the research team and set to be released publicly at DEF CON 34 (https://github.com/bl4cksku11/entraith) — enabling device code attacks, token renewal, email and app inspection, token exfiltration, and persistence generation from a single interface.

  This talk is not about a single CVE or a niche edge case. It is a comprehensive offensive research presentation covering the full spectrum of attack techniques against Microsoft Entra ID — from the very first foothold to deeply rooted, multi-vector persistence that survives incident response.

  We will demonstrate, live, how an attacker moves through every phase of an Entra ID compromise:

  Initial access and token theft: We begin with Device Code Flow phishing — abusing Microsoft's own authentication protocol to harvest valid tokens without ever touching a password. We then escalate with PowerShell process hijacking (GrabTokenAzureAD) to intercept live session tokens, and culminate with Sliver BOF-based extraction of the Primary Refresh Token (PRT) directly from memory — stealthy, silent, and anti-malware evasive. These three techniques alone demonstrate that MFA and Conditional Access are not the guarantees organizations believe them to be.

  Bypassing Intune and MFA: Purpose-built tools developed during this research extract local PRTs, bypass Intune device compliance controls, and circumvent MFA enforcement at the token layer — not through social engineering, but through direct abuse of Microsoft's identity platform mechanics. Attackers operating from non-enrolled or non-compliant devices can achieve full Tenant access that Conditional Access policies are designed to prevent.

  Persistence across 10 vectors: Once inside, we deploy a "Survival Kit" of 10 chained persistence mechanisms — camouflaged App Registrations, long-lived secrets and certificates, backdoor accounts, strategic role assignments, privileged group inheritance, admin mailbox delegation, inbox-rule-based exfiltration of credentials and MFA codes, OAuth consent grants to external applications, and Temporary Access Pass generation for on-demand MFA bypass. No single blue team action — password reset, MFA enforcement, device wipe — removes all of them simultaneously.

  Entraith — built by us, released at DEF CON: The centerpiece of this talk is Entraith (https://github.com/bl4cksku11/entraith), an offensive framework designed and built from the ground up by the speakers specifically for this research. Entraith is not a wrapper around existing tools — it is original work, developed over months of hands-on red team engagements against real Microsoft 365 environments. It unifies every technique demonstrated in this talk into a single operator interface: device code phishing, PRT extraction from memory, Intune compliance bypass, token renewal and exfiltration, email and application enumeration, and the full 10-step persistence deployment chain. DEF CON 34 will be the moment Entraith is released to the public. Attendees will be among the first to access the tool, its documentation, and the full methodology behind it — making this talk a genuine first-look at original research with immediate real-world impact.

  Why this matters for Cloud Village: The Microsoft 365 and Entra ID ecosystem is the identity backbone of the majority of enterprise organizations worldwide. The techniques presented here are not theoretical — they are being used by real threat actors today. Defenders in the room will leave with concrete detection opportunities and an understanding of exactly which log sources and control gaps allow this attack chain to succeed undetected. Red teamers will leave with a working tool and a fully documented methodology. This talk delivers both offensive depth and defensive utility in a single session, and the live demos ensure no attendee has to take our word for it.

  ResearchersElzer Pineda PA·Jose Manuel Rivas PA

  Presented at Red Team Village — DEF CON 2026

  View official session Entraith on GitHub
• August 9, 2026 Talk AppSec & Supply Chain AppSec Village — DEF CON 2026 Filipi Pires BR From commit to compromise: securing the full pipeline with AI-assisted remediation

  Most vulnerabilities aren't found because teams lack tools they persist because the cost of fixing them is too high. A finding without a concrete, context-aware fix is just noise developers learn to ignore. This talk walks through three real-world attack scenarios a secret leaked into git history, a compromised dependency in a supply chain attack, and an injection vulnerability introduced in a PR and shows how each is detected, mapped to OWASP Top 10:2025, and automatically remediated using AI running entirely on local infrastructure. No source code leaves the machine. The AI receives the vulnerable code block, CWE identifier, CVSS score, and OWASP category and returns a fix that is specific, correct, and ready to apply. Attendees leave with a working open-source tool and a new mental model for what AppSec remediation can look like.

  ResearchersFilipi Pires BR

  Presented at AppSec Village — DEF CON 2026

  View official session
• August 9, 2026 Talk Offensive Security La Villa Hacker — DEF CON 2026 Andrés Sabas MX Tu foco IoT nunca tuvo modelo de amenazas

  Compraste una foco IoT. La conectaste a tu WiFi. Le pusiste una contraseña. ¿Quién más le puede mandar órdenes? ¿Quién verifica que el firmware que descarga es legítimo? En esta charla cuento la auditoría completa de Magic Home, la app de Zengge instalada en bombillas, tiras LED y switches que se venden bajo una docena de marcas blancas. No vengo a enumerar bugs. Vengo a mostrar tres preguntas básicas que el ecosistema entero, móvil, nube y radio, no puede responder una sola raíz: nunca hubo modelo de amenazas en el ciclo del producto. Uso firmware ya publicado en internet, scripts caseros y un decompilador. Cero hardware comprado. Una demo grabada de por qué nada detendría a un atacante. Una demo en vivo de cuánto se puede ver del fleet sin tocar a nadie más.

  ResearchersAndrés Sabas MX

  Presented at La Villa Hacker — DEF CON 2026

  View official session
• August 9, 2026 Talk AI Security La Villa Hacker — DEF CON 2026 Juan Carlos Sequeira Piedra CR·Jose Urena CR Red Robin: Don't Hack For Me, Hack With Me

  Desde 2025, y especialmente con la llegada de modelos de IA cada vez más capaces, el panorama de la ciberseguridad ha visto una explosión de herramientas de hacking impulsadas por IA. Prometen explotación automatizada. El objetivo de muchas de estas herramientas es ser 100% autónomas, con el hacker fuera del loop. Pero esto es seguridad: sistemas en producción, datos sensibles. Una operación real no mide solo la tecnología, también a las personas y los procedimientos, cómo detectan y cómo responden. Sin contexto y sin alguien que responda por cada decisión, automatizar no alcanza.

  Esta charla presenta Red Robin, un compañero de IA por línea de comandos hecho específicamente para red teamers. Red Robin se integra con tu framework de Command-and-Control (C2), observa la actividad de tus beacons a medida que ocurre, y te da guía contextual en tiempo real anclada a tus objetivos operativos. Nunca actúa por su cuenta. No hackea por vos. Piensa con vos.

  Esta charla cubre el panorama actual de hacking humano-vs-IA, la filosofía detrás del modelo de compañero, una inmersión técnica en la arquitectura de Red Robin, una demostración en vivo de Red Robin construyendo situational awareness dentro de una sesión de C2 activa, y una discusión a futuro sobre dónde debería encajar la asistencia de IA en el futuro de la seguridad ofensiva.

  ResearchersJuan Carlos Sequeira Piedra CR·Jose Urena CR

  Presented at La Villa Hacker — DEF CON 2026

  Speaker page (Juan Sequeira) Speaker page (Jose Urena)
• August 8, 2026 Talk AI Security La Villa Hacker — DEF CON 2026 Lenin Alevski MX·Asher Davila MX Keep the Model on Course: Agentic LLM Workflows for Reversing

  Malware analysts are wiring LLMs into their reversing workflows today. The MCP integrations for IDA Pro, Ghidra, radare2, and Binary Ninja are public, and analysts use them to rename functions, triage samples, and hunt vulnerabilities. Adoption has outpaced any measured account of where these tools hold up and where they mislead, and running an agent against one sample can burn an enormous number of tokens on work the model repeats for every binary.

  We built an agentic malware analysis framework on ADK (Agent Development Kit) that connects LLMs to disassemblers through MCP, wrapped in per-language skills for C/C++, C#, Go, Android, etc. The ADK agent loop drives the analysis: pulling functions, following xrefs, reading decompiled output, making notes, and issuing disassembler commands on its own. Each skill supplies the runtime context once, so the model spends its budget on the sample. Everything runs in Docker containers for isolation and reproducibility.

  We tested it across multiple models on real samples and scored every result against ground truth. Triage and function renaming on stripped binaries work well when the agent is steered. Crypto identification and CVE matching fail in ways that read as authoritative. The analyst is the captain, the model the crew. We ship the containers, skills, and prompt templates so attendees can reproduce similar workflows efortlessly.

  ResearchersLenin Alevski MX·Asher Davila MX

  Presented at La Villa Hacker — DEF CON 2026

  Speaker page (Lenin Alevski) Speaker page (Asher Davila)
• August 8, 2026 Talk Offensive Security Red Team Village — DEF CON 2026 Lenin Alevski MX Red Teaming Kubernetes: From App-Level CVEs to Full Cluster Takeover

  Kubernetes is the de facto operating system of the cloud, and more and more organizations are running their workloads on Kubernetes. While Kubernetes offers many benefits, it also introduces new security risks, such as cluster misconfiguration, leaked credentials, cryptojacking, container escapes, and vulnerable clusters.

  In this workshop, attendees will learn how to attack Kubernetes clusters by simulating a real-world adversary exploiting one of the most recent vulnerabilities in the ecosystem: IngressNightmare (CVE-2025-1974). Participants will practice exfiltrating service account tokens and credentials, performing lateral movement, escalating privileges by targeting common applications deployed in Kubernetes environments, and ultimately compromising the entire cluster.

  ResearchersLenin Alevski MX

  Presented at Red Team Village — DEF CON 2026

  View official session
• August 8, 2026 Workshop AI Security Recon Village — DEF CON 2026 Lenin Alevski MX Vibe Coding Your Way to a Fully Functional Open-Source Intelligence Platform

  Everyone is talking about vibe coding, but most examples stop at TODO apps, landing pages, or simple CRUD applications. This workshop goes far beyond that.

  Participants will use modern AI coding agents to build a real-world Open-Source Intelligence (OSINT) platform from scratch designed to aggregate and visualize intelligence data on an interactive 3D globe.

  Rather than simply watching a demo, attendees will actively build the application alongside the instructor while learning the mindset, prompting techniques, and engineering workflow required to successfully collaborate with AI coding assistants.

  The workshop begins with the fundamentals of vibe coding: what it actually is, why it works, where it fails, and how to communicate effectively with AI agents. Participants will learn practical prompting strategies, iterative development techniques, and methods for breaking large software systems into manageable tasks that AI can successfully implement.

  From there, we'll progressively build an intelligence dashboard capable of consuming and visualizing multiple real-time OSINT sources, including flight tracking, satellite positions, earthquake feeds, maritime traffic, and other publicly available intelligence signals. Along the way, participants will generate modern user interfaces, integrate external APIs, debug AI-generated code, and learn when to trust the model and when not to.

  Rather than treating AI as a code generator, this workshop teaches attendees how to use AI as an engineering partner capable of dramatically increasing development velocity while still maintaining high-quality software.

  By the end of the workshop, every participant will have:

  1. A working open-source intelligence platform running locally

  2. A practical workflow for building complex software using AI coding agents

  3. A reusable prompting framework applicable to future projects

  4. Experience integrating multiple public OSINT data sources

  5. An understanding of the strengths and limitations of AI-assisted software development

  6. A roadmap for continuing to extend the platform after the workshop

  The workshop is based on the open-source project: https://github.com/Alevsk/respondent-community

  Workshop Outline (150 Minutes)

  Part 1 — Introduction to Vibe Coding (20 min)

  • What vibe coding actually means
  • Mental models for collaborating with AI
  • Selecting the right AI model for the task
  • Understanding agentic coding workflows
  • Common mistakes and misconceptions

  ---

  Part 2 — Prompt Engineering for Software Development (25 min)

  • Structuring prompts that produce maintainable code
  • Breaking large systems into incremental tasks
  • Designing software through conversation
  • Iterating instead of regenerating
  • Knowing when to intervene manually

  ---

  Part 3 — Building the Intelligence Platform (75 min)

  Participants will build the platform together while learning how to:

  • Generate a modern frontend
  • Create a real-time 3D globe visualization
  • Integrate multiple public OSINT APIs
  • Display flights, satellites, earthquakes, ships, and additional intelligence feeds
  • Build reusable UI components with AI
  • Debug AI-generated code
  • Refactor and improve generated implementations
  • Continue extending the application using AI as a development partner

  ---

  Part 4 — Lessons Learned & Advanced Techniques (20 min)

  • What AI does well
  • Where AI still struggles
  • Managing technical debt
  • Building larger projects with AI
  • Cost optimization strategies
  • Recommended workflows and tooling
  • Future directions for AI-assisted engineering

  ResearchersLenin Alevski MX

  Presented at Recon Village — DEF CON 2026

  View official session
• August 8, 2026 Talk Offensive Security DEF CON 34 Dan Borgogno AR·Javier Bernardo AR The Glass Perimeter: Systematic Bypasses in Biometric Frameworks and the Rise of Synthetic Identity

  Identity is the new perimeter, and biometrics are its supposed gatekeepers. But what happens when the gatekeepers are blind to the reality they consume? We spent more than 6 months deconstructing the biometric "Root of Trust" across every top-tier framework we could find, solutions relied upon by the world's largest banks and providers worth billions. The result: A 100% bypass rate. From high-fidelity physical spoofs to a first-of-its-kind Cross-PID buffer hijack against integrated anti-tamper SDKs, we prove that even multi-million dollar "fortresses" can be reduced to client-side theater.

  This talk is a technical journey through the guts of the mobile media pipeline, exposing how synthetic identities are manufactured at scale to bypass RASP, Kernel-level integrity, and AI models. When the "Master Key" is just another line of code an attacker can hook, the risk isn't just a bug, it's a systemic failure affecting millions of users. Customers are losing wealth, companies are buying illusions, and the glass perimeter is shattering.

  ResearchersDan Borgogno AR·Javier Bernardo AR

  Presented at DEF CON 34

  View official session
• August 8, 2026 Talk Incident Response La Villa Hacker — DEF CON 2026 Eduardo Chavarro Ovalle CO Meet XEntry Team: A powerful threat actor abusing Bitlocker for ransomware

  Our GERT team was notified about a company affected by a previously unknown threat actor identified as "XEntry Team". Initially, the customer identified this threat as a hijacked LogMeIn session that allowed the attackers to activate BitLocker and encrypt the disks of every system synchronizing to the AD, but our analysis confirmed a 2 months intrusion, using a mix of clever techniques and lack of monitoring, that allowed attacker to configure a bridge to the infrastructure for:

  • Recognition
  • Critical assets identification
  • Exfiltration
  • Persistance
  • ransomware deployment
  • Other not so funny stuff

  During this session we will present to the audience the investigation results, the techniques implemented by the threat actor, the scope of the attack from beginning to the end, and the analysis and attempts to recover affected systems.

  ResearchersEduardo Chavarro Ovalle CO

  Presented at La Villa Hacker — DEF CON 2026

  View official session
• August 8, 2026 Talk Malware & Reversing Malware Village — DEF CON 2026 Nelson Rafael Colón Merán DO Haetae: An Agent to Takedown North Korean C2 Servers

  In this talk, we introduce Haetae, an agent designed to profile, identify, and exploit C2 frameworks used by North Korean malware.

  Haetae supports both automated and interactive modes, allowing analysts to map and understand adversary infrastructure with different levels of control. It is built around a flexible rule-based system, enabling users to extend and refine detections as new patterns and frameworks emerge.

  In this first release, we will walk through real-world cases where Haetae was used to identify and take down infrastructure associated with Mach-O Man and POWerful Armadillo.

  We will also release a safe emulator of both malware C2 servers, allowing researchers and newcomers to experiment with Haetae in realistic, controlled environments without risk.

  This is a highly technical talk but can be enjoyed by both beginners and seasoned threat hunters alike.

  ResearchersNelson Rafael Colón Merán DO

  Presented at Malware Village — DEF CON 2026

  View official session
• August 8, 2026 Talk AI Security La Villa Hacker — DEF CON 2026 Simón Correa CO·Michael Rivera CO Descubrimiento Industrializado de CVEs con Agentes de IA

  ¿De verdad puede la IA encontrar vulnerabilidades reales y nuevas a gran escala? La seguridad de aplicaciones es nuestra disciplina, y somos una CNA de tipo investigador, entre las primeras del CNA Scorecard. Montamos una cadena de producción y la soltamos sobre 36 proyectos de código abierto. Hasta ahora, y contando: 539 hallazgos revisados, depurados hasta 8 CVEs públicos y 30 candidatos a CVE.

  La charla recorre la cadena completa: cómo entrenamos agentes que recorren el código, cómo un modelo de machine learning detecta los fragmentos sospechosos, cómo los hackers y researchers validan cada candidato y qué números deja cada etapa del filtrado.

  Nuestra meta: acortar el tiempo de detección a escala sin sacrificar exactitud, para que la ventana de exposición (detección + remediación) sea menor que el tiempo de explotación. Te llevas un método replicable para convertir LLMs impredecibles en una fábrica seria de CVEs, con las métricas de precisión que la sostienen.

  ResearchersSimón Correa CO·Michael Rivera CO

  Presented at La Villa Hacker — DEF CON 2026

  Speaker page (Simón Correa) Speaker page (Michael Rivera)
• August 8, 2026 Talk AI Security ICS Village — DEF CON 2026 Asher Davila MX·Malav Vyas Give an AI Industrial Protocol Tools and Watch What It Destroys

  Every major ICS attack of the last decade succeeded not because of software vulnerabilities, but because industrial protocols were built to trust any packet on the wire. The village has read the incident reports. What it doesn't have is a way to replay them against its own infrastructure to learn what its detections actually catch and what they miss.

  We present mrhOTshOT, an open-source framework that emulates history's most destructive ICS attacks across the complete kill chain, reconstructed from publicly available incident analyses. Not just the OT payload the full chain: Windows initial access with real CVEs, lateral movement to engineering workstations, protocol-native process manipulation, and persistent physical impact. Every emulation generates wire traffic consistent with publicly documented behavior on the correct industrial protocol for that attack family.

  The framework spans a wide range of industrial protocols across ten distributed PLCs, each simulating the real-world process that protocol actually controls: a heating district controller for Modbus, a safety instrumented system for TriStation, a centrifuge cascade for S7comm. Nothing runs on a generic simulated tank with ten protocols bolted on.

  We also introduce the Agentic Attack Emulation Framework: every protocol action is exposed as a callable tool, orchestrated by an LLM agent that reads live process state and composes attack sequences on the fly. No hardcoded playbook, you decide. This is what AI-assisted ICS attack composition looks like, and defenders need to understand it before they meet it in the wild.

  The talk closes with a live demo: three centrifuges destroyed in real time while the operator HMI, deceived by an S7 rootkit, shows normal operation throughout, until it doesn't.

  ResearchersAsher Davila MX·Malav Vyas

  Presented at ICS Village — DEF CON 2026

  View official session
• August 8, 2026 Talk Incident Response La Villa Hacker — DEF CON 2026 Ashley Hiram MX Execution of modern techniques to start/improve your career in Incident Response

  Practicaremos algunas técnicas modernas que utilizamos en equipos globales de Respuesta a Incidentes para procesar y analizar evidencias digitales durante ataques cibernéticos, y responder con conclusiones acertadas en un lapso muy corto.

  ResearchersAshley Hiram MX

  Presented at La Villa Hacker — DEF CON 2026

  View official session
• August 8, 2026 Talk Malware & Reversing DEF CON 34 Federico Kirschbaum AR·Gaston Aznarez AR Root From Kilometers Away: Ubiquiti AirMax RCE

  You don't realize it until you see them; they're everywhere. From Wireless ISP links to the frontline of modern warfare. But nobody found anything? The devices behind those links are Ubiquiti AirMAX: critical infrastructure on a 17-year-old Linux kernel and a custom 802.11 extension built on "security by obscurity." So we took it apart. This talk covers our reverse engineering of the AirMAX protocol, AirOS, and the kernel modules behind this proprietary mode. It rides on 802.11 Information Elements that look encrypted, but we'll show why they aren't. What we found: two critical vulnerabilities (CVE-2026-21639, CVE-2026-21638) across airMAX AC, airMAX M, airFiber, and GigaBeam, over 50 devices. These are the bugs from the movies: Over-The-Air, unauthenticated, kernel-privilege RCE. No network access, just line of sight. They affect every AirMAX device ever shipped. We disclosed them through Ubiquiti's bug bounty program. The bugs were rated "Adjacent", except adjacent here means kilometers away. The same hardware can be turned around and pointed at the problem: we'll repurpose these devices as recon tools and release open-source software to locate AirMAX networks in the wild. This talk is about our journey, our tooling, and the state of security.

  ResearchersFederico Kirschbaum AR·Gaston Aznarez AR

  Presented at DEF CON 34

  View official session
• August 8, 2026 Workshop AI Security Red Team Village — DEF CON 2026 Mario Lobo CO AIMARU C2: The New Era of LotL (MCP AI-Driven C2)

  AImaru C2 is the evolution of a concept sparked by an unsettling question: What happens when an attacker can execute advanced Living off the Land (LotL) attacks without being an expert? While traditional LotL requires a specialist to navigate complex environments, AImaru C2 changes the game. Evolving from a PoC into a sophisticated Red Team framework, it automates expert-level tradecraft by subverting Anthropic's Model Context Protocol (MCP). By weaponizing MCP clients, it transforms a standard context protocol into a functional, autonomous multi-level RAT that natively "speaks" PowerShell.

  AImaru C2 was born out of a critical necessity to ground cybersecurity in reality. While the industry often distracts itself with over-engineered narratives and hype-driven threats that ignore actual ecosystem data, a far more pragmatic danger has been hiding in plain sight. This framework addresses the unsettling evolution of the "Expert Bottleneck": What happens when the sophisticated LotL tactics, once reserved for elite nation-state actors, are democratized through automation? Named after the mythical serpent that glides unseen between the underworld and the land of the living, AImaru C2 represents the shift from manual expertise to autonomous execution. It is no longer just a proof of concept; it is a full-scale Red Team framework designed to expose how easily an unskilled attacker can now weaponize authorized system protocols to orchestrate advanced, data-driven breaches that traditional defenses simply aren't looking for.

  To understand its impact, context is key. For years, LotL attacks have been the dominant tactic in ransomware campaigns. Instead of deploying traditional malware, threat actors abuse legitimate OS tools—such as PowerShell, WMI, or Certutil—to operate under the radar. In 2025, 82% of successful breaches were fileless or malwareless, with PowerShell ranking as the second most utilized TTP by groups like Black Basta, Royal, and LockBit.

  This robust framework moves beyond experimental concepts, consolidating itself as a production-ready offensive platform. AImaru C2 is engineered with a modular architecture that prioritizes cryptographic integrity, operational security, and intelligent automation.

  Core Framework Capabilities:

  • Model Context Protocol (MCP) RAT Engine: A paradigm shift in command and control that subverts Anthropic's MCP from a standard context-sharing tool into a functional, undetectable Remote Access Trojan (RAT), tunneling all C2 traffic through legitimate LLM API communication.
  • PowerShell Client Builder: An automated obfuscation engine featuring deep variable randomization and structural mutation to break static signatures.
  • AMSI Bypass Generator: Dynamic generation of multi-stage bypass scripts, utilizing diverse techniques to neutralize the Antimalware Scan Interface in real-time.
  • Real-time Monitoring: Advanced telemetry for live command execution tracking and granular session management.
  • Cryptographic Isolation: Implementation of per-client encryption keys derived via HKDF-SHA256, ensuring that the compromise of one beacon does not jeopardize the entire fleet.
  • RBAC: Native Role-Based Access Control, strictly segregating permissions between Admins, Users, and Viewers.
  • Complete Audit Logging: Forensic-grade command history with millisecond-accurate timestamps for post-operation deconfliction.

  By utilizing a multi-tier decision logic, the system dynamically selects and rewrites payloads based on the specific operational objective. This allows the C2 to scale its complexity in real-time—transitioning from basic PowerShell structures and stealthy WMI manipulation to the strategic abuse of LOLBins for high-complexity tasks. This adaptive approach ensures that every command utilizes the most effective legitimate system tool, successfully evading environment-specific defenses by blending into the target's unique operational noise.

  Tactic Description: In this interactive session, attendees will sit down with the developer to operate the AImaru C2 framework in a controlled lab environment. Participants will experience the "Natural Language Intent" paradigm shift firsthand, moving away from manual syntax to intent-based exploitation.

  Deployment: Deploy a lightweight MCP-Client on a target Windows machine and establish a beacon back to the AI-C2 controller.

  Intent-Based Recon: Issue high-level natural language prompts (e.g., "Find sensitive PDF files and identify lateral movement paths") and observe the LLM selecting and generating the appropriate PowerShell/WMI code in real-time.

  Automated Evasion: Trigger the Adaptive AMSI Bypass module, witnessing how the framework re-writes its own memory-resident payloads to evade active defenses.

  ResearchersMario Lobo CO

  Presented at Red Team Village — DEF CON 2026

  View official session Project on GitHub
• August 8, 2026 Talk Malware & Reversing La Villa Hacker — DEF CON 2026 María Isabel Manjarrez MX Malware Inc.: Cybercrime-as-a-Service y la economía del cibercrimen moderno

  El cibercrimen moderno dejó atrás el modelo del atacante aislado. Hoy existe un ecosistema modular y altamente especializado donde servicios como Malware-as-a-Service (MaaS), Ransomware-as-a-Service (RaaS), Pay-Per-Install (PPI), Initial Access Brokers (IAB) o Phishing-as-a-Service (PhaaS) permiten que distintos actores colaboren dentro de una cadena de suministro criminal distribuida.

  En esta charla exploraremos cómo el malware adoptó principios de industrialización similares a los de la industria tecnológica legítima: afiliados, automatización, revenue sharing, soporte técnico y servicios bajo demanda. A través de casos reales revisaremos cómo funciona esta economía clandestina, cómo se monetizan distintas etapas de un ataque y por qué entender el modelo de negocio detrás del malware es clave para comprender las amenazas actuales.

  ResearchersMaría Isabel Manjarrez MX

  Presented at La Villa Hacker — DEF CON 2026

  View official session
• August 8, 2026 Talk Offensive Security La Villa Hacker — DEF CON 2026 Gerardo Mejia CR When Cash Runs on Windows: A Red Team Look at ATM Attack Surfaces

  Muchos cajeros automáticos siguen siendo, en esencia, computadoras Windows especializadas. Algunos son modernos, otros llevan años operando con sistemas heredados, configuraciones antiguas y controles que nunca fueron revisados a profundidad. Están en gasolineras, supermercados y esquinas, conectados a infraestructura crítica donde una mala decisión de hardening, segmentación o administración remota puede convertirse en un riesgo real. Nadie los toca. Nadie los cuestiona. Hay dinero adentro. Esta charla resume hallazgos de ejercicios reales de pentesting sobre ATMs realizados en América Latina.

  Veremos cómo se ve la seguridad ATM en producción: qué controles suelen estar presentes, cuáles fallan de formas inesperadas y cómo debilidades aparentemente menores pueden encadenarse hasta comprometer infraestructura crítica. Entre los hallazgos analizaremos fallas en controles de endpoint, hardening, segmentación, administración remota y exposición de servicios legacy. También veremos qué controles resistieron, por qué lo hicieron y qué diferencia a un ATM correctamente endurecido de uno que puede convertirse en punto de entrada hacia infraestructura crítica bancaria. Más que mostrar "cómo atacar un cajero", esta charla busca crear conciencia sobre la importancia de endurecer y monitorizar estos equipos.

  ResearchersGerardo Mejia CR

  Presented at La Villa Hacker — DEF CON 2026

  View official session
• August 8, 2026 Talk Offensive Security La Villa Hacker — DEF CON 2026 Fernando Mengali BR De la Explotación de Buffer Overflows al C2 Industrial: Infectando y Controlando RTUs con Malware

  En esta charla práctica y directa al punto, se demostrará cómo una vulnerabilidad de buffer overflow puede ser explotada para comprometer entornos SCADA, evolucionando desde la ejecución de shellcode multietapa hasta el establecimiento de canales de comunicación C2 y el control remoto de sistemas industriales críticos. A través de demostraciones técnicas, se analizará cómo una vulnerabilidad aparentemente aislada puede convertirse en una cadena de ataque capaz de afectar procesos industriales, provocar interrupciones operativas e incluso derivar en escenarios de blackout.

  ResearchersFernando Mengali BR

  Presented at La Villa Hacker — DEF CON 2026

  View official session