Duration: 15 minutes
Bio
Spencer McIntyre is the Director of Exploit Development at VulnCheck. Spencer is an offensive security-oriented R&D professional; avid open-source contributor and Python enthusiast. Previously consulted across healthcare, energy, and manufacturing sectors.
Description
This talk announces the Initial Access Community, an open repository that drops a new, working exploit every single week, targeting the unauthenticated, no-interaction bugs in VPNs, appliances, and self-hosted services that get you in the door. The real engine is go-exploit, an open-source Go framework that makes every exploit small, portable, and consistent: it standardizes target verification, version scanning, and exploitation into a single self-contained binary with no runtime dependencies to drag around. Because output is proxy-aware and emits clean key-value or JSON, the same exploit that pops one box scales straight into mass scanning and slots into your existing tooling. Catch the shell and move on. Come see how to grab a fresh exploit, spin up a vulnerable lab, and fire it off in minutes.
Duration: 45 minutes
Bio
AJ Hammond is a Lead Security Engineer on the Red Team at Praetorian, where he specializes in long-period fully covert offensive cybersecurity operations. AJ is also the lead social engineer at Praetorian and the Active Directory subject matter expert.
Description
"Ignore previous instructions" earns a refusal in under a second now. Single-turn jailbreaks are dead weight on an assessment, and most LLM testing still stops there.
Augustus is Praetorian's open-source LLM vulnerability scanner. Single Go binary, Apache 2.0, 210+ probes across 47 attack categories, 28 providers, no Python to fight with on a client box. It proxies through Burp and speaks REST, so the proprietary endpoints clients actually deploy are in scope.
The probe library isn't what makes it interesting. The multi-turn engine is. It runs three LLMs: an attacker drafting messages, the target under test, and a judge scoring progress toward your objective every turn. Four strategies, four different sets of tradecraft. Crescendo escalates so gradually that no single message trips a filter. GOAT stacks seven techniques behind Chain-of-Attack-Thought reasoning. Hydra erases refused turns so the target's history stays clean and it never escalates defensively. Mischievous User drops the adversarial framing entirely and probes like a curious customer. Attacker, target, and judge can each be a different model from a different provider, so a local Ollama attacker can drive the whole scan and keep it off your API bill.
I'll walk through the pipeline (probe, buff, generator, detector), buff chaining as an evasion multiplier, how each strategy performed against hardened targets, and a live demo of Hydra backtracking through a refusal. I'll also cover the failure modes nobody blogs about: judge false negatives, attacker parse failures, and why a vague goal wastes an entire scan.
You'll leave with working configs and a methodology that holds up on a real engagement.
Duration: 45 minutes
Bio
Nic Losby is a Principal Consultant on the Mandiant Offensive Security Red Team. Over seven years of professional security consulting, Nic has performed dozens of pentests and red-team operations. He is a subject-matter expert in internals, password auditing, embedded devices, and offensive tooling.
Description
This talk will cover yet another multi-year journey (May 2020 to Feb 2026) from dream to countless failures and ending with the project finally working and in some cases, beating an RTX 5090’s hashing performance for $60 while using a twentieth of the power.
The talk opens with a short history of FPGAs through the lens of password cracking and crypto attacks. From there the talk digs into the hole left by the discontinued ZTEX 1.15y boards, and the design decisions that made maintenance and porting the John the Ripper (JtR) modules to any other FPGA so painful no one succeeded. Next, why other attempts to reinvent the JtR modules failed and the missing framework which made the entire project not only possible but downright easy to maintain and port. This will be followed by a discussion of the development of a hashcat bridge module along with the resulting PR. Finally, a look into why this project is not a magic solution for every hash type and discuss future efforts for wider algorithm support.
Duration: 45 minutes
Bio
Garrett is a Senior Security Researcher at SpecterOps specializing in Windows tradecraft and attack path development. His research focuses on Active Directory security and enterprise management infrastructure.
Description
In this talk, I plan to publicly disclose five new SCCM vulnerabilities that were reported to Microsoft. I'll show how to dump distribution point content files...unauthenticated. Demonstrate command injection in CMPivot queries as well as some "read-only" command execution. Finally, the known speak-to-your-manager management point relay gets an upgrade with two SQL injection bugs (in 2026) that lead to complete hierarchy takeover.
Duration: 45 minutes
Bios
Mike Pinch: Mike Pinch is CTO of Security Risk Advisors.
Greg Stachura: Greg Stachura is a Senior Manager at Security Risk Advisors.
Description
Attackers don’t always need custom malware to succeed. Often, the same administrative tools that keep an organization running can help an adversary establish persistence or expand a compromise. For defenders, that creates a difficult challenge: how do you distinguish legitimate activity from malicious behavior without burying your SOC in alerts?
Single-event detection rules can flag suspicious actions, but often lack the context needed to separate routine administration from an intrusion. User and Entity Behavior Analytics (UEBA) and risk-based alerting offer additional context, yet can still generate excessive false positives or overlook subtle activity that never crosses a risk threshold. The result: more noise for analysts, and opportunities for attackers to go unnoticed.
This session introduces the Statistical Funnel, an approach that combines targeted statistical models with AI-assisted investigation to narrow the field before digging deeper. Statistical models identify unusual activity, such as scheduled task or service creation, relative to an asset’s history, a specific time window, or patterns across the organization. Those signals surface a smaller set of hosts and users for deeper investigation, helping focus analyst attention where it matters most.
Duration: 45 minutes
Bios
Skyler Knecht: Skyler Knecht is a Senior Security Consultant at SpecterOps. He performs security assessments for large organizations, focuses on initial access research, and contributes to open-source tooling and vulnerability research. Skyler has presented at conferences including DEF CON and BSides.
Kevin Clark: Kevin Clark is a Security Consultant with TrustedSec and a Red Team Instructor. He has a background in software development, penetration testing, and offensive operations, and regularly delivers training and conference presentations.
Description
During Red Team operations, operators often lack visibility into the client's landscape, leading to unforeseen issues with stage-one and stage-zero payloads. Although these payloads provide several features that make operating easier, those features increase the frequency with which these issues occur. SOCKS proxies, on the other hand, require minimal code to work, permitting operators to turn most applications into a SOCKS proxy. Therefore, all operators should be able to establish a SOCKS proxy in any environment if the primary payload is inoperable. To demonstrate this, the speakers will leverage a tool called Messenger to turn several applications into SOCKS proxies, including those written in common frameworks such as .NET and Electron.
Duration: 45 minutes
Bio
Brett Hawkins has over 10 years of experience in offensive security, helping lead teams at JPMorgan Chase, Mandiant, and IBM X-Force Red. At Armadin, he leads the offensive network security service lines and helps build offensive capabilities used in complex environments.
Description
MLOKit is an open-source C# toolkit that can be used to attack MLOps platforms by taking advantage of available REST APIs. The tool allows a user to specify an attack module along with valid credentials for a target MLOps platform. MLOKit currently supports Azure ML, Amazon SageMaker, Google Cloud Vertex AI, MLFlow, BigML, and Palantir AIP. Attack modules include reconnaissance, training data theft, model theft, model poisoning, and notebook attacks.
This release introduces 9 new modules that significantly expand MLOKit's attack surface coverage. New capabilities include enumerating ML compute instances and their SSH configurations, adding attacker SSH keys to compute instances for persistent access, creating malicious training jobs for arbitrary code execution, harvesting credentials from training job variables and log files, downloading Docker build files from ML environments that may contain embedded secrets, and executing code directly within JupyterLab notebook instances via WebSocket connections. These new modules target Azure ML and Amazon SageMaker, extending MLOKit's capabilities from data and model theft into compute access, credential harvesting, and code execution within ML training infrastructure.
Duration: 15 minutes
Bio
Vina Ta is the founder of VeeSafe Technology, focused on translating security and compliance into practical action. She has over seven years of cybersecurity experience, including supporting industrial and manufacturing environments, and specializes in making complex concepts actionable for technical and non-technical audiences.
Description
Most offensive security professionals learn in traditional IT environments where scanning, patching, and testing are routine. Industrial and operational technology (OT) environments play by a different set of rules.
In OT, uptime, safety, and production often matter more than confidentiality. A vulnerability scanner, aggressive assessment, or well-intentioned security recommendation can create real-world operational consequences that don't exist in enterprise IT.
This talk explores the mindset shift required when moving from offensive security in IT environments to industrial environments. We'll cover common assumptions that break down in OT, why operations teams often push back on security recommendations, and lessons learned from working with manufacturing organizations. Attendees will leave with a practical understanding of how to approach OT engagements more effectively and avoid becoming the security team everyone dreads seeing walk into the plant.
Whether you're interested in red teaming, ethical hacking, or industrial security, this session will highlight why understanding the environment is just as important as understanding the technology.
