Learning paths

Talks learning path

18 resources, ordered. Work down the list — each stage assumes the one above it.

Presentations / Talks

A route through the library rather than a dump of it. Items are drawn from the full Talks collection, filtered to teaching material (news items are excluded) and ordered within each stage by depth signals — whether the piece carries code, how substantial it is, and what readers actually open.

1

Start here

5 resources

Orientation and first principles — what the bug class is and how it behaves.

  1. AI Agent Security Masterclass: Attacking and Defending Autonomous AI Systems - DEF CON Training
    27 min readtraining.defcon.org2026
    Training covers attacking and defending autonomous AI agents, focusing on building secure agents, exploiting weaknesses, and implementing defenses against prompt injection, excessive agency, tool misuse, and MCP-based supply chain attacks. Participants will learn about AI agent architecture, RAG security, threat modeling AI systems, and practical red-teaming exercises against vulnerable agents, culminating in defense strategies like least privilege, sandboxing, and robust prompt controls.
  2. Hacking Android and IOT Apps by Example - DEF CON Training LV 2026
    9 min readtraining.defcon.org2026
    Workshop slides from DEF CON LV 2026 detail hands-on techniques for hacking Android and IoT applications, focusing on the OWASP Mobile Security Testing Guide (MSTG) and MASVS. Participants will learn static and dynamic analysis methods, advanced instrumentation with Frida, Objection, and radare2, and how to bypass common security controls like certificate pinning and root detection. The training covers identifying vulnerabilities through app analysis without device access, efficient exploitation, and leveraging open-source tools for mobile security testing.
  3. DEF CON AppSec Village talk on malicious AI coding-agent skills
    defcon.social2026
    AI coding agents, while boosting productivity, also pose security risks as malicious skills can be installed. Patrick Smyth of Chainguard demonstrated how these skills can exfiltrate data through fake update checks or fetch instructions from remote URLs, allowing for silent manipulation. He also showcased Chainguard's hardening pipeline, which automatically identifies risky elements like obfuscated commands and excessive permissions in skills before they are used by agents.
  4. BSidesSF 2026: AI-Powered AppSec: 10x your security team without scaling headcount (Anshuman Bhartiya)
    youtube.com2026
    Anshuman Bhartiya's presentation at BSidesSF 2026, "AI-Powered AppSec," focuses on leveraging artificial intelligence to significantly enhance application security team efficiency. The core idea is to achieve a tenfold increase in output ("10x") without requiring proportional growth in personnel. This approach addresses the overwhelming volume of vulnerabilities, pull request reviews, and manual tasks that often burden security teams, suggesting AI as a solution to overcome these challenges.
  5. DEF CON 34: Beyond the Ceremony, the 2026 Passkey Attack Surface
    media.defcon.org2026
    This DEF CON 34 talk, "Beyond the Ceremony, the 2026 Passkey Attack Surface," explores potential vulnerabilities and attack vectors for passkeys. It delves into the evolving landscape of passwordless authentication and anticipates challenges and security considerations for passkey adoption leading up to and around 2026. The presentation aims to highlight the critical areas where security needs to be strengthened as passkey technology becomes more widespread.
2

Build depth

5 resources

Real testing methodology, tooling, and writeups that show the work.

  1. Write Once, Shell Everywhere - Turning Arbitrary File Writes into RCE (DEF CON Bug Bounty Village)
    20 min readethiack.com2026
    Library cataloging techniques for escalating arbitrary file writes to remote code execution, focusing on modern, stripped-down environments. It categorizes existing methods by operating system, language runtime, and framework, evaluating their usefulness in contemporary security engagements. Specific techniques discussed include leveraging `/etc/ld.so.preload` and identifying vulnerabilities within language runtime deserialization processes, offering practical insights for discovering and exploiting file write primitives.
  2. Hacking AI customer service agents
    18 min readintigriti.com2026
    Talk from DEF CON 34 detailing methods for weaponizing AI customer service agents. Researchers demonstrate how attackers can exploit AI chatbots via email, including bypassing multi-factor authentication, spoofing sender addresses using multiple `From` headers and `Sender` headers, and tricking victims into sending signed emails to the AI agent. These techniques can lead to unauthorized actions and data exfiltration, potentially yielding significant bug bounty rewards.
  3. DEF CON 34 BugBounty: AI Nightmare - Hacking at 0 Hour (Pedro Paniago)
    2 min readcpebuddy.com2026
    Talk from DEF CON 34 detailing "AI Nightmare" bug bounty hunting, where scaffolding prompts, including CLAUDE.md and credentials, enabled exploitation of auth bypasses, IDORs, business logic flaws, and account takeovers. Standout critical vulnerabilities included a gVisor sandbox escape and zero-click ATO via rogue ADFS certificates. The presentation also covered AI's limitations, autonomous hunting tips, and the "zero hour era" where CVE weaponization accelerates, exemplified by AI reversing a TeamCity CVE in five minutes.
  4. DEF CON 34: C2YA, inside the adversary's inbox (Vitaly Simonovich)
    2 min readcpebuddy.com2026
    Talk from DEF CON 34 details Vitaly Simonovich's use of LLM coding agents (Claude Code, Codex, Gemini) to uncover 250 flaws in five open-source C2 frameworks: Havoc, Mythic, Silver, Covenant, and Adaptic. Findings include unauthenticated AES key broadcasts in Havoc, IV reuse enabling cryptanalysis, and RCE via the Roslyn-compiled "message transform" field in Covenant. The research highlights the necessity for defenders to actively assess adversary tooling, emphasizing that vulnerabilities can be exploited by both attackers and defenders.
  5. DEF CON 34: Finding prompt injection vulns via mutated prompt generation
    3 min readcpebuddy.com2026
    Tool is the Mutated Prompt Generator (MPG) framework, a divergent search tool for discovering prompt injection vulnerabilities. It automates vulnerability discovery by mutating payloads to test patch generalization beyond specific exploits. MPG addresses indirect prompt injection attacks and leverages parser leniency while employing specific mutation operators and a subtree viability scoring algorithm to identify exploitable vulnerabilities.
3

Go deep

8 resources

Novel research, edge cases, and the techniques that push the class forward.

  1. Can AI do novel security research? Meet the HTTP Terminator
    29 min readportswigger.net2026
    Tool: The HTTP Terminator, an autonomous security research system, explores the frontier of AI-driven bug hunting. It generates novel HTTP desync triggers, gadgets, and exploits, demonstrating novel attack techniques that compromised live systems. The tool analyzes discovery chains, enabling the transformation of human expertise into autonomous weapons, and identifies research areas beyond current AI capabilities, including undisclosed recon techniques and new attack classes. Presented at Black Hat USA 2026 and DEF CON 34, the HTTP Terminator's open-source release is intended to help other researchers adopt this automated approach.
  2. HTTP Terminator: an AI system proposed a new desync attack class (DEF CON)
    3 min readtheinjection.dev2026
    Library for automating HTTP desync attack research, the HTTP Terminator, was presented at DEF CON 34. This autonomous system, built by James Kettle, ideates, evaluates, weaponizes, and cascades novel desync techniques like Shared-Parser Confusion and dual-matching Content-Length headers. It targets web servers and proxies that misinterpret request boundaries, having identified a zero-day in Apache Traffic Server and impacting over 200 sites during testing. The source code and updated tools like HTTP Request Smuggler are available for defenders.
  3. DEF CON 34: A billion user blast radius, owning ChatGPT's secure sandbox
    2 min readcpebuddy.com2026
    Talk from DEF CON 34 detailing a full command-and-control chain against ChatGPT. The demonstration exploited the "?q=" URL parameter for prompt injection, unsupervised Python code execution via spreadsheet parsing, reasoning injection, and a covert channel using JFrog Artifactory lockouts to exfiltrate and manipulate data across a billion users. The research challenged assumptions about private chats remaining private and sandboxed code execution's inability to exfiltrate data.
  4. DEF CON 34: Beyond the Ceremony - the 2026 passkey attack surface (Matteo Giordano)
    divert.stream2026
    Matteo Giordano at DEF CON 34 explored the attack surface of passkeys by 2026, moving beyond the typical ceremony. His research revealed potential vulnerabilities and pathways for exploitation. The presentation aimed to highlight the evolving security landscape of passkeys and how they might be targeted in the future.
  5. DEF CON 34: From Fuzzer Noise to Zend Engine 0day
    media.defcon.org2026
    This DEF CON 34 talk, "From Fuzzer Noise to Zend Engine 0day," details how researchers leveraged fuzzer output to discover a critical Zend Engine 0-day vulnerability. The presentation likely covered the process of analyzing fuzzing data, identifying patterns indicative of a bug, and ultimately exploiting it within the Zend Engine. The content suggests a technical deep-dive into vulnerability research methodologies.
  6. Exploiting The Off-Chain Ecosystem In Web3 Bug Bounty (DEF CON 33)
    youtube.com2026
    At DEF CON 33's Bug Bounty Village, Bruno Halltari highlighted the often-neglected "off-chain ecosystem" as a significant attack vector in Web3 applications. His presentation focused on how vulnerabilities outside the blockchain itself, such as in front-ends, APIs, and off-chain infrastructure, can be exploited to compromise Web3 systems. This approach expands the traditional bug bounty scope beyond smart contract audits, urging researchers to consider the entire application stack for potential security weaknesses.
  7. BlueHat Asia 2026: Weaponizing client-side chains to shatter cloud identity and AI boundaries
    youtube.com2026
    The BlueHat Asia 2026 conference will feature a presentation titled "Weaponizing client-side chains to shatter cloud identity and AI boundaries." This session will explore advanced techniques for exploiting vulnerabilities within client-side application components. The goal is to demonstrate how these weaknesses can be leveraged to bypass existing security measures, ultimately compromising cloud identity systems and potentially impacting artificial intelligence models. The presentation aims to highlight novel attack vectors and the critical need for robust defenses against sophisticated client-side threats.
  8. Análisis avanzado de JavaScript para Bug Bounty (BountyMagicCon)
    youtube.com2026
    Francisco Neves, creator of JXScout, will present an advanced JavaScript analysis approach for bug bounty hunting at BountyMagicCon. The talk, titled "Análisis avanzado de JavaScript para Bug Bounty," focuses on uncovering the "magic behind the bounties" through in-depth techniques.