Black Hat Europe 2025 | Flaw And Order: Finding The Needle In The Haystack Of CodeQL Using LLMs

Running CodeQL's built-in queries on Redis gave me over 6,800 potential issues. Doable, maybe. But when I tried FFmpeg, I got over 51,000. That's way too much for me. And how many of those are real vulnerabilities? Probably around 0.01%. The sheer number of false positives makes static code analysis impractical - who wants to manually sift through tens of thousands of results just to find a few actual security flaws? To fix this, we built an open-source tool that fuses CodeQL with an LLM-driven agent. This agent autonomously navigates the code, running targeted queries to extract only the relevant context. On top of that, we introduced Guided Questioning, an advanced reasoning technique that keeps the LLM focused, improving accuracy even for complex vulnerabilities. Using this approach, we reduced false positives by up to 97% and uncovered more than a dozen real-world security issues in Linux, Apache, FFmpeg, Bullet3, Libvips, libretro, Linenoise, and other widely used open-source projects. By: Simcha Kosman | Senior Security Researcher, Cyberark https://ift.tt/NSzZIgc

source https://www.youtube.com/watch?v=gcsIUqb6s_8

Black Hat Europe 2025 | A crash course in revealing insecure blind spots for DoS & DDoS

Domain Controllers (DCs) are organizations' crown jewels. A successful Denial-of-Service (DoS) attack against them can terminate authentication processes and cause widespread disruption. Our previous LdapNightmare research - the first public pre-auth DC DoS exploit for CVE-2024-49113, revealed that DCs can be turned into LDAP clients by communicating with their NetLogon RPC server. These clients could then be crashed by a single invalid value they receive. This taught us that remotely triggered client code is a blind spot that overtrusts. Eager to find other blind spots in servers on DCs, we asked - what will make server code overtrust? abstraction layers! We realized that although common server code nowadays mostly mitigates classic server risks, that's maybe untrue in case it's transport-agnostic, uses heavy abstractions, and focuses mostly on the application's logic. Starting by targeting remotely triggered LDAP client code, we found a vulnerability that denies service from DCs, or alternatively can be exploited to manipulate them to join a DDoS botnet attack. Then, we moved on to target Windows' most common transport-agnostic wrapped server code - RPC functions. By exploiting security gaps in RPC bindings, we developed novel techniques allowing to hammer a single RPC server tens of thousands of times simultaneously from a single system, far surpassing standard concurrency limits! And WOW- this armed us beyond our expectations, with vulnerabilities crashing any form of Windows, both servers and endpoints! Our blind spot hypothesis turned out to be the reality. In this talk, we'll present "Win-DoS" - A set of tools exploiting 30 DoS vulnerabilities we discovered in Domain Controllers and Windows endpoints. Most vulnerabilities do not require any authentication, and one even allows not only to crash, but also to effortlessly initiate a botnet harnessing the upload rates and vast resources of any public DCs to participate in DDoS attacks. By: Or Yair | Security Research Team Lead, SafeBreach Shahak Morag | Security Researcher https://ift.tt/u0jKlfE

source https://www.youtube.com/watch?v=VV-R9JQFRkw

Black Hat Europe 2025 | Unveiling System Management Mode Memory Corruption Vulnerability Via Fuzzing

System Management Mode (SMM) is an operating mode introduced by the x86 processor to handle critical hardware events and chipset errors. SMM applications, designed to run in this mode, operate at a high privilege level (known as Ring -2, which is even higher than the kernel mode, Ring 0). With the high privilege, SMM applications have almost unlimited access to system resources. However, vendors commonly adopt memory-unsafe programming languages, such as C and C++, to develop SMM applications, making them prone to memory corruption vulnerabilities. Once compromised, the attacker may gain complete control over the system. This intrinsic feature makes SMM applications a very attractive target for attackers. While SMM applications play a crucial role in the foundation of low-level system software, applying efficient and effective fuzzing to them is a very challenging and complex task. In this talk, we present the first systematic SMM application fuzzing framework specifically designed to detect memory corruption vulnerabilities in closed-source SMM applications. We observe that the SMM application, as part of the UEFI firmware, is supposed to run in a UEFI runtime environment. Without such an environment, SMM applications cannot be correctly initialized and executed. As such, we will present all the technical details related to an all-in-one solution for SMM application fuzzing. Our framework offers a fully featured UEFI runtime environment. With such an environment, we ensure that fuzzing does not result in early crashes and a high number of false positives. Additionally, we present the details behind a universal fuzzing harness for successful fuzzing campaigns. The fuzzing harness contains an interface grouping and a memory access interception mechanism to infer the input semantics, such that it can explore the deep logic of SMM applications. Our framework has already proven its impact: in our experiments, we identified a total of 38 new vulnerabilities in firmware from nine well-known vendors. We will share the technical insights behind these discoveries and walk through several real-world case studies that highlight the power and versatility of our approach. By: Jianqiang Wang | Dr.-Ing., Max Planck Institute for Security and Privacy https://ift.tt/GSgiRT2

source https://www.youtube.com/watch?v=OXxSc4-sn9Q

Black Hat Stories | Ari Herbert-Voss, CEO and Founder of RunSybil

In episode 5, Ari Herbert-Voss, Founder and CEO of RunSybil, talks about a first Black Hat experience — the scale, the technical depth, the community that's both welcoming and sharp. After multiple years of attendance, each event continues to showcase new technologies and evolving approaches, while the community stays the same. That's what keeps practitioners coming back. From emerging trends like offensive AR to the accelerating pace of offensive security, Black Hat highlights where the industry is headed. Peer-reviewed research. Hands-on training from practitioners deploying techniques in live environments. A community that is open to new voices.

source https://www.youtube.com/watch?v=zkSzT3G_Gz4

SecTor 2025 | Grand Finale: Cutting Through the Cyber Noise

Join our Review Board members for a powerful closing session that distills the essential cybersecurity insights from this year's conference. This dynamic panel will synthesize key takeaways from the Briefings program and forecast emerging trends that security professionals should have on their radar. Leave SecTor with clarity on what truly matters in today's complex threat landscape. Opheliar Chan | Chapter Co-Lead, OWASP Toronto Dave Millier | CSO, Quick Intelligence Maryna Neprosta | Review Board, SecTor Tom Tran | Senior Manager of Offensive Security, Government of Ontario https://ift.tt/zmNcfv8

source https://www.youtube.com/watch?v=queScDSGzxk

SecTor 2025 | Chasing Shadows: Chronicles of Counter-Intelligence from the Citizen Lab

For over twenty years, the University of Toronto's Citizen Lab has pioneered investigations into digital security and human rights—from exposing state cyber espionage to uncovering the global spread of mercenary spyware targeting journalists, activists, and human rights defenders. Drawing from my latest book, Chasing Shadows, I will recount how our mission to conduct "counter-intelligence for civil society" revealed surveillance around the inner circle of murdered Washington Post journalist Jamal Khashoggi and uncovered domestic espionage campaigns across Mexico, Spain, Hungary, Poland, Thailand, El Salvador, and most recently, Italy. As our small team disarmed cyber mercenaries and helped improve the digital security of billions, we, too, became targets—caught in the same sinister crosshairs as those we sought to protect. I will also look ahead to the future of our mission and the rising challenges of AI-enabled subversion, Dark PR, and advertising intelligence, and how the kind of public-interest research the Lab has championed is now under threat from a growing tide of despotism and authoritarianism. By: Ron Deibert | Director, The Citizen Lab, Professor of Political Science, University of Toronto https://ift.tt/yHF7MPC

source https://www.youtube.com/watch?v=QNJgYIpKWWE

SecTor 2025 | The Good, the Bad, and the Ugly: Hacking 3 Cloud Providers with 1 Vulnerability

Join us for an inside look at how leading cloud providers architect their environments, and the anatomy of a container escape vulnerability in the wild. Our goal is to learn how to build stronger guardrails in the cloud by examining the flaws and misconfigurations we were able to exploit in each environment. As AI workloads migrate to the cloud, Cloud Providers are rapidly evolving their GPU offerings. These multi-tenant environments are often built on the NVIDIA Container Toolkit, the industry-standard framework for running GPU-based containerized apps. In this talk, we will show you how a single vulnerability in this fundamental framework impacted the entire cloud ecosystem – and how each environment handled a brand-new 0-day vulnerability. We'll walk through our discovery of a container escape vulnerability in this foundational layer of GPU infrastructure, and its real-life implications across 3 different cloud providers: Azure, DigitalOcean, and Replicate. Each case study began with a standard customer workload running our exploit – but the outcomes varied widely. One led to a minor impact; another with lateral movement that triggered blue teamers; and one resulted in complete service takeover. The differing outcomes didn't stem from the vulnerability itself; they stemmed from varying service architectures and security best practices. We'll analyze and contrast these implementations to demonstrate how a well-isolated environment can be resilient even against 0-day attacks! By: Hillai Ben-Sasson | Security Researcher, Wiz Nir Ohfeld | Head of Vulnerability Research, Wiz https://ift.tt/pXBaFGb

source https://www.youtube.com/watch?v=O49EgRnu4VU

SecTor 2025 | Security is Easier Before PCB Assembly: Easy Threat Modeling for Hardware

Most threat modeling ignores hardware — but hardware problems can be impossible to fix when products have left the factory. The industry has spent decades refining threat modeling processes so they're approachable, organized, and useful; however most of this was done with software security in mind. Three leading experts have performed a threat model of the OpenWRT One. We'll share our complete results, a case study threat modeling document, and our process. We chose it because it's open and attendees may be familiar with it, but also because the scenario mirrors real threat modeling: you don't have to reverse out all the details. Whether we're dealing with IoT/OT devices, hardware security modules, multi-tenant cloud hardware, or specialized compute accelerators, we've seen when and how hardware-specific threats come into play. When is hardware in scope? When is it someone else's problem? When and how do we decide if it is just an acceptable risk? We'll explain when, why and how your next model should consider hardware threats, even if you don't think you have hardware to worry about or you think it's out of scope. We'll call out a number of assumptions you should keep in mind and share the process for you to assess mixed hardware/software systems yourself. Attendees will learn how to develop a better understanding of what hardware you're already working with, what can go wrong with it, and what you can do about it. Hopefully this, combined with a fully worked example of how that all comes together, will help you do a good job of incorporating hardware concerns into your threat model to make long term product security easier. By: Eric Evenchick | Co-Founder and Managing Partner, Tetrel Security Joe FitzPatrick | Trainer and Researcher, SecuringHardware.com Adam Shostack | President, Shostack + Associates https://ift.tt/5vCmArq

source https://www.youtube.com/watch?v=W5p6zGpjdLY

SecTor 2025 | Scaling the AppSec Program Without Scaling Security Headcount

The ability to scale application security programs, including vulnerability triage and remediation with bots has been proven. This session will apply a flavor of GenAI, enhanced with proprietary data accumulated through years of very large-scale security deliveries and focus on how to implement the bot(s), what scales can be achieved, and the cost savings and results. By: Mario Lauande Lacroix | Senior Security Manager, Accenture Will Yeager | Security Consulting Manager, Accenture https://ift.tt/ch3Am9y

source https://www.youtube.com/watch?v=QNPH4kh4dQY

SecTor 2025 | Invoking Gemini for Workspace Agents with Simple Google Calendar Invite

Over the past two years, we have witnessed the emergence of a new class of attacks against LLM-powered systems known as Promptware. Promptware refers to prompts (in the form of text, images, or audio samples) engineered to exploit LLMs at inference time to perform malicious activities within the application context. While a growing body of research has already warned about a potential shift in the threat landscape posed to applications, Promptware has often been perceived as impractical and exotic due to the presumption that crafting such prompts requires specialized expertise in adversarial machine learning, a cluster of GPUs, and white-box access. This talk will shatter this misconception forever. In this talk, we introduce a new variant of Promptware called Targeted Promptware Attacks. In these attacks, an attacker invites a victim to a Google Calendar meeting whose subject contains an indirect prompt injection. By doing so, the attacker hijacks the application context, invokes its integrated agents, and exploits their permission to perform malicious activities. We demonstrate 15 different exploitations of agent hijacking targeting the three most widely used Gemini for Workspace assistants: the web interface (www.gemini.google.com), the mobile application (Gemini for Mobile), and Google Assistant (which is powered by Gemini), which runs with OS permissions on Android devices. We show that by sending a user an invitation for a meeting (or an email or sharing a Google Doc), attackers could hijack Gemini's agents and exploit their tools to: Generate toxic content, perform spamming and phishing, delete a victim's calendar events, remotely control a victim's home appliances (connected windows, boiler, and lights), video stream a victim via Zoom, exfiltrate emails and calendar events, geolocate a victim, and launch a worm that tarets Gemini for Workspace clients. Our demonstrations show that Promptware is capable to perform (1) inter-agent lateral movement (triggering malicious activity between different Gemini agents), and (2) inter-device lateral movement, escaping the boundaries of Gemini and leveraging applications installed on a victim's smartphone to perform malicious activities with physical outcomes (e.g., activating the boiler and lights or opening a window in a victim's apartment). Finally, we assess the risk posed to end users using a dedicated threat analysis and risk assessment framework we developed. Our findings indicate that 73% of the identified risks are classified as high-critical, requiring the deployment of immediate mitigations. By: Or Yair | Security Researcher, SafeBreach Stav Cohen | PhD Student, Technion Ben Nassi | Ramat Gan, Technion https://ift.tt/YHvcmWX

source https://www.youtube.com/watch?v=KVUniT5H4Rk

SecTor 2025 | Rethinking Phishing Detection in the Age of AI and Disinformation

Phishing is no longer just a technical problem; it is a cognitive one. Classifiers that rely on dynamic features such as sentiment, urgency, or message length are highly vulnerable to concept drift. Attackers adapt quickly, and with the help of large language models, they can now craft highly convincing phishing messages that evade traditional detection systems. Many of the signals we once relied on are no longer dependable because they also appear in legitimate communication. In response, there is a growing shift toward static features, especially URL-based analysis. Elements like domain entropy or subdomain structure are harder for attackers to modify without breaking the link and tend to remain stable over time. However, static models often lack transparency. Why was the link flagged? What pattern triggered the detection? Without clear explanations, users are left in the dark and trust in the system erodes. This Briefing explores how to move beyond surface-level detection. Drawing on recent research in adversarial machine learning, social engineering modeling, and cognitive psychology, we will present a classifier design that integrates manipulation scoring, concept drift monitoring, and explainability from the ground up. Attendees will gain insight into how phishing tactics evolve and how to design defenses that adapt to them while staying aligned with human behavior. By: Michel Hebert | Industry Research, Practice Lead, Info-Tech Research Group https://ift.tt/eDCdFOz

source https://www.youtube.com/watch?v=nAbyzHJivfo