Black Hat Asia 2026 | Graph-Aware LLM for Windows Logon with a Closed-Loop Guarded Detection Agent

Because Windows Event Logs were never originally designed for detecting unauthorized logons, traces of attacks are easily buried in a massive amount of noise. It is also inherently difficult to create reliable signatures for suspicious log entries in Windows Event Logs, and research on analysis methods has been ongoing for many years. In recent years, the use of LLMs for log analysis has advanced; however, in real-world investigations, log sizes often exceed hundreds of gigabytes. In such cases, prompts quickly become too large, making it impractical to apply LLMs directly. In addition, hallucinated explanations and lack of reproducibility remain key challenges when using LLMs in security operations. In this Briefing, we will present a practical, production-ready framework that combines graph analytics with LLM agents to accurately detect suspicious logons. Concretely, we compress logs into graph information by constructing an authentication graph of users and hosts from Windows Event Logs. This makes it possible to reduce the data to a realistic size that can actually be fed to an LLM. On top of that, a closed-loop detection agent autonomously iterates the cycle of generating search queries to a database → executing the queries → evaluating the results → exploring further. Through this loop, it detects signs such as concentrations of service tickets, cross-host logons by privileged accounts, remote service access, and suspicious chains of logons. The outcome of the analysis is presented as an incident severity level, an evidence timeline, and an attack scenario summary, automatically providing information that can be directly used in real incident investigations. Our approach aggregates millions of events down to a few dozen suspicious logons within minutes, and elevates LLM usage in DFIR into a form that is auditable, reproducible, and operationally viable. We will release an open-source tool that implements this method so that analysts can apply it to real-world incident analysis. Shusei Tomonaga | CTO, JPCERT/CC https://ift.tt/B2jZ1tL

source https://www.youtube.com/watch?v=7QVcpQA_j_I

Black Hat Asia 2026 | Social Media Manipulation Wargaming for Cyberliteracy and Research

LLM-powered bots are influencing social media around the world, amplifying the effects of misinformation and disinformation. Detecting if content is being promoted or manipulated by such bots is difficult, both for casual observers of social media platforms as well as for platform operators faced with continuous "cat and mouse" development by malicious adversaries. To address both challenges, we produced "Capture the Narrative", the world's first CTF-style massively-multiplayer wargame for social media manipulation. Over four weeks in 2025, 288 students formed 70+ teams from Australian universities and competed in a multiplayer competition to manipulate an in-house social media platform we called "Legit Social". They were aiming to swing a simulated election in a fictional country, a task which they ultimately completed by moving the electorate by 1.8 percentage points. In this Briefing, I will introduce the competition, as well as discuss its outputs, which include the associated awareness campaign as well as a collaboratively developed dataset made up of survey data, developed bot software, and recorded bot activity for the thousands of bots producing millions of artificial posts to amplify in-game misinformation. Hammond Pearce | Senior Lecturer, UNSW (University of New South Wales) https://ift.tt/A7mkBlS

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

Black Hat Asia 2026 | When 3.5 Billion Strangers Can Exploit Your WhatsApp Devices

WhatsApp evolved from a simple text-messaging app serving thousands of users in 2009 to the world's largest messaging platform with 3.5 billion users and 30+ complex message types, but its original "anyone can contact anyone" architecture remained unchanged. This creates a systemic security risk where users' attack surface grows linearly with both features and users. In this talk, we will demonstrate how attackers abuse WhatsApp's unrestricted access architecture for attacks ranging from mundane spam to APT zero-click remote code execution exploits. We will present two novel attack classes responsibly disclosed to Meta: Class I attacks that exploit nonconsensual message delivery, and Class II attacks that violate user privacy by extracting metadata directly from WhatsApp servers. We'll reveal our new device fingerprinting techniques leveraging implementation side-channels in E2EE and messages, release an open source WhatsApp research tool demonstrating these attacks, and provide live exploitation demos against test accounts. Moving to defenses, we will reveal why current mitigations fail because they treat individual symptoms rather than the architectural root cause, creating a perpetual whack-a-mole cycle where each new feature potentially introduces new vulnerability classes. To thoroughly solve this issue, we will present a viable architectural solution that dramatically reduces the attack surface while maintaining strong privacy guarantees. Most critically, this solution makes the attack surface fixed with respect to WhatsApp's expansion in users and features, transforming security from reactive patching to secure by default. Tal Be'ery | CTO, Co-Founder, Zengo (acquired by eToro) https://ift.tt/KIcYoCM

source https://www.youtube.com/watch?v=FXV0LWPcB64

Black Hat Asia 2026 | Shedding LIGHT on Real-World Attacks on Cloudless IoT Devices

The rapidly expanding Internet of Things (IoT) landscape is shifting toward cloudless architectures, removing reliance on centralized cloud services but exposing devices directly to the internet and increasing their vulnerability to cyberattacks. Our research revealed an unexpected pattern of substantial Tor network traffic targeting cloudless IoT devices, suggesting that attackers are using Tor to anonymously exploit undisclosed vulnerabilities (possibly obtained from underground markets). To delve deeper into this phenomenon, we developed TORCHLIGHT, a tool designed to detect both known and unknown threats targeting cloudless IoT devices by analyzing Tor traffic. TORCHLIGHT filters traffic via specific IP patterns, strategically deploys virtual private server (VPS) nodes for cost-effective detection, and uses a chain-of-thought (CoT) process with large language models (LLMs) for accurate threat identification. Our results are significant: for the first time, we have demonstrated that attackers are indeed using Tor to conceal their identities while targeting cloudless IoT devices. Over a period of 12 months, TORCHLIGHT analyzed 26 TB of traffic, revealing 45 vulnerabilities, including 29 zero-day exploits with 25 CVE-IDs assigned (5 CRITICAL, 3 HIGH, 16 MEDIUM, and 1 LOW) and an estimated value of approximately $312,000. These vulnerabilities affect around 12.71 million devices across 148 countries, exposing them to severe risks such as information disclosure, authentication bypass, and arbitrary command execution. The findings have attracted significant attention, sparking widespread discussion in cybersecurity circles, reaching the top 25 on Hacker News, and generating over 190,000 views. Yumingzhi Pan | Ph.D. Student, Southeast University Zhen Ling | Professor, Southeast University Yue Zhang | Professor, Shandong University Hongze Wang | Ph.D. Student, Southeast University Guangchi Liu | Professor, Southeast University Junzhou Luo | Professor, Southeast University https://ift.tt/QDxGp3A

source https://www.youtube.com/watch?v=UM-ej4hO9_U

Black Hat Asia 2026 | Exploiting DFIR Agents Through Adversarial Manipulation

In recent years, Digital Forensics and Incident Response (DFIR) tools have increasingly adopted Large Language Models (LLMs) to enhance automation, analysis, and reporting. Prominent examples include Velociraptor's MCP integration and Timesketch's AI Summary feature. This study empirically demonstrates that attackers can exploit prompt injection through boundary perturbation of structured data—a form previously considered resistant to manipulation. Importantly, this issue is not specific to any single tool; rather, it represents a broader class of risks that emerges whenever DFIR tools are integrated into autonomous LLM agents. By embedding malicious instructions into routine forensic artifacts such as logs and scheduled tasks, adversaries can cause DFIR LLM agents to misinterpret benign data as instructions, leading to three outcomes: Hide, Mislead, and Exploit. To the best of my knowledge, this is the first work to demonstrate structured-data injection attacks in LLM-integrated DFIR environments. The study also proposes practical defense-in-depth countermeasures, including enforcing least privilege, mandating strict structured output validation, and maintaining human-in-the-loop verification to ensure the reliability and safety of automated DFIR workflows. This Briefing aims to provide organizations advancing DFIR automation with LLM agents a foundation for rethinking, at the design level, how much autonomy should be granted to such agents and where human oversight must remain integral. Yusuke Nakajima | Security Analyst, NTTDATA https://ift.tt/RftH9kP

source https://www.youtube.com/watch?v=sK8omfWUMaM

Black Hat Asia 2026 | WhisperPair: A Security Analysis of Google Fast Pair

Google Fast Pair has promised "one-tap" Bluetooth onboarding and seamless account synchronisation across phones, laptops and tablets, since 2017. In practice, it has quietly become the default pairing path for modern earbuds, headphones and speakers across the Android ecosystem. Users trust that once an accessory is bonded, it will not suddenly attach to somebody else's phone without explicit consent. This Briefing shows that this trust was misplaced. We will present WhisperPair, a family of attacks that let a nearby adversary hijack Fast Pair compatible accessories that are not in pairing mode, seize audio, activate microphones, and silently attach the victim's device to the attacker's Google account for long term location tracking and stalking. The trick is simple but devastating: although the Fast Pair specification requires accessories to reject unauthorised pairing requests, a wide range of chipsets and vendors fail to enforce this in practice. Using only commodity hardware and standard Bluetooth stacks, we evaluated 25 commercial earbuds, headphones and speakers from 16 brands, covering what we believe to be all major audio manufacturers currently supporting Fast Pair. Most of them could be hijacked in under 15 seconds, and every vulnerable model that supported Google's Find Hub extension allowed covert account binding and stalking until factory reset. The Briefing walks through the attack in live demos, dissects what went wrong in Google's compliance chain, and releases a practical test harness that defenders can run against their own products. We will close with our proposed solution: IntentPair, a drop in protocol hardening that cryptographically binds user intent into Fast Pair without sacrificing usability. Our findings will show how a small usability "add-on" can introduce large-scale security and privacy risks for hundreds of millions of users when intent is not cryptographically bound, and how to address this to avoid such mass-scale problems. For further information about this work, please visit https://whisperpair.eu/ Seppe Wyns | PhD Student, DistriNet, KU Leuven Sayon Duttagupta | Scientific Researcher, COSIC, KU Leuven Nikola Antonijević | PhD Student, COSIC, KU Leuven Dave Singelée | Associate Professor, DistriNet - Group T, KU Leuven Bart Preneel | Professor, COSIC, KU Leuven https://ift.tt/x5sb0SN

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

Black Hat Asia 2026 | Cast Attack: A New Threat Posed by Ghost Bits in Java

In modern security defense systems, input validation and data integrity checks are the core to preventing attacks. However, a commonly overlooked source of vulnerabilities has long lurked in the code, not due to complex logic errors, but because of "ghost bits" silently erased during type conversion. This presentation reveals a novel attack technique called Cast Attack, which originates from data loss during Java's type casting process. In this talk, we will demonstrate how Cast Attack can be used to bypass defenses such as WAFs, as well as introduce four major attack surfaces: privilege/access bypass, arbitrary file read, SMTP injection, and XSS. Affected vendors include, but are not limited to Oracle, Spring, Eclipse, Apache, Atlassian, JetBrains, and others. The impact of Cast Attack far exceeds expectations. It not only challenges current input validation mechanisms but also provides attackers with a low-cost, stealthy attack vector that can cause unforeseen security vulnerabilities in critical systems. Could this become the next widely exploited attack technique? In this session, we will uncover this hidden threat together. Xinyu Bai | Security Researcher, Zhihui Chen | Security Engineer, Alibaba Cloud Zongzheng Zheng | Independent Researcher, University of New South Wales https://ift.tt/sZxYatH

source https://www.youtube.com/watch?v=HhbLr4LKIl0

Black Hat Asia 2026 | Practical Attacks Against Smartphone Boot ROMs

Boot ROMs are an immutable component of any hardware platform, and vulnerabilities in them can compromise the entire boot chain. This talk aims to outline the impact of vulnerabilities in the USB interface of smartphone Boot ROMs at a low level, demonstrating how a smartphone's entire ecosystem can be manipulated from a single vulnerability. Using example targets from two different manufacturers, this talk aims to outline the impact of code execution vulnerabilities in multiple boot stages of a smartphone, providing practical demonstrations of secure boot bypass on mobile devices, on-device fuzzing, and decryption of protected firmware images. Christopher Wade | Staff Engineer https://ift.tt/QhSE2uv

source https://www.youtube.com/watch?v=0hItwQVp8a4

Black Hat Asia 2026 | Subverting Screen Trust via State Disruption and ONE-WAY Flooding

As core components of graphics and input architecture, SurfaceFlinger and InputDispatcher share critical responsibilities in physical screen handling. Although they are tightly integrated in the system architecture and interact frequently via the Binder driver, surface composition and input processing are fundamentally independent workflows. This seemingly unremarkable premise exposes a unique attack surface: by forcing SurfaceFlinger's scheduler into mishandling VSYNC signals, malware can fully disable the device's UI protections. Android's ongoing tapjacking mitigations have made zero-permission exploitation extremely difficult over the past 8 years. Our talk aims to break this deadlock by exploiting multiple vulnerabilities to successfully attack the latest Android 15 devices and threaten nearly all downstream OEM vendors! This research spans SurfaceFlinger, SystemServer and WMShell, and also covers several critical core services. Additionally, we will introduce a universal exploitation technique that bypasses standard defenses—even when all protection mechanisms are functioning as intended—by exploiting design flaws in the Binder transaction mechanism. Overall, the attack chain enables zero-permission bypass of privileged window control logic, rendering most UI protections, including TRUSTED_OVERLAY, ineffective. These findings have earned over $42,000 in vulnerability rewards, with one vulnerability remaining unpatched since 2023. Beyond theoretical feasibility, this research will present the POC on production devices to validate its practical reliability and stability. Additionally, we will develop a fully weaponized version and simulate real-world malware targeting the system permission controller. This enables privilege escalation that exceeds conventional expectations, without any user awareness. WeiMin Cheng | Independent Researcher, Zhihan Lin | Security Engineer, Chengdu Royal Security Technology Co., Ltd. Sheng Cao | Mobile Security Researcher, Huazhong University of Science and Technology Songzhou Shi | Security Researcher, LSPosed Team https://ift.tt/aDCH02W

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

Black Hat Asia 2026 | Systematically Exploring and Exploiting DNS Silent Vulnerabilities

Domain names function as human-readable identifiers on the Internet, with characters serving as their essential building blocks. However, since the initial specification of domain names in 1983, the security implications of handling special characters within the domain name resolution process have remained largely overlooked. In this work, we conducted the first systematic study of special character handling logic in DNS, reviewing DNS RFCs and analyzing 31 widely-used DNS software implementations through source code review and gray-box testing. Our systematic analysis reveals two new DNS logic vulnerabilities arising from inconsistencies and silent handling behaviors, leading to two classes of attacks (four variants) that affect all DNS roles, including stub resolvers, forwarders, recursive resolvers, and authoritative nameservers. We name them the SHAR attack. Attackers can exploit these vulnerabilities to launch DNS cache poisoning and load balancing disruption attacks. Through comprehensive experiments, we validated the impact on the real world. All 31 tested mainstream DNS software implementations are vulnerable to SHAR. Notably, attackers can seize control of domain names, even the entire TLD or deceive victim resolvers to return invalid responses for legitimate queries, resulting in a persistent DoS effect. The SHAR attack can also enhance 10/13 well-known off-path DNS cache poisoning attacks (2002–2025). To further determine the impact in the wild, we test all DNS-related roles, including mainstream Wi-Fi routers, router OSes, public DNS services, table open DNS resolvers, Root servers, TLD servers, SLD servers, and domain names. The results show that the SHAR attack affects all tested Wi-Fi routers, router OSes, and public DNS services. In addition, we identified that over 12.5M domain names are also vulnerable to the SHAR attack. Following the best practice of responsible disclosure, we have reported these vulnerabilities to all affected vendors. Fasheng Miao | Master Student, Tsinghua University Xiang Li | Associate Professor, Nankai University Changqing An | Associate Researcher, Tsinghua University Jilong Wang | Professor, Tsinghua University https://ift.tt/Y5SJER7

source https://www.youtube.com/watch?v=2Euva5ZT-cA

Black Hat Asia 2026 | More JVM Memory Shells: JVM Memory Shell Auto Searching Program

A Java memory shell is a fileless backdoor that resides entirely in JVM memory, leaving no trace on disk. Attackers exploit code execution vulnerabilities—such as ScriptEngine injection or deserialization flaws—to use Java reflection to replace legitimate objects in web frameworks with malicious classes. Once implanted, specially crafted HTTP requests (mimicking normal traffic) trigger arbitrary command execution within the JVM, with results exfiltrated via standard HTTP responses. This stealthy technique blends seamlessly into legitimate traffic and bypasses firewalls that only allow ports 80/443, rendering traditional reverse shells ineffective. Over the past eight years, common variants have included Tomcat Filter, Tomcat Listener, and Spring Controller memory shells—all dynamically injected at runtime. However, the discovery of new types has largely stalled in recent years, relying almost exclusively on manual source code audits. We have developed an automated framework for discovering Java memory shells, integrating SAST (Static Application Security Testing), Java Agent–based hooking, JVM runtime memory introspection, and AIpowered PoC generation and validation capabilities. This framework dramatically accelerates the discovery of novel memory shells: in a very short time, it expanded the number of known Spring memory shell variants from just 2 to 9. Moreover, it is adaptable to any Java web framework for uncovering new memory shell techniques, significantly enhancing the efficiency of Java memory shell research and surpassing years of manual efforts. Litong Wan | Cyber Security Engineer, Alibaba Holding - Risk & Security Dept Fanghai Yu | Independent Security Researcher, Yang Jing | Cyber Security Engineer, Alibaba Holding - Risk & Security Dept Dongyan Zhang | Senior Security Engineer, Alibaba Holding - Risk & Security Dept Huan Zeng | Senior Security Engineer, Alibaba Holding - Risk & Security Dept https://ift.tt/FJb7ihN

source https://www.youtube.com/watch?v=YIuqIDm1mfk