CAMLIS 2019
DAY ONE
Bio: Aleatha Parker-Wood
Aleatha Parker-Wood is the Machine Learning and Algorithmic Privacy lead at Humu, a company dedicated to making work better for everyone everywhere. Prior to Humu, she was a Sr. Principal Research Engineer and manager in the Center for Advanced Machine Learning at Symantec, where her team did original research and contributed machine learning to numerous Symantec products including SEP 14, Email Security.cloud, Norton Core, phishing page detection, and more.
She holds multiple security-related patents, and serves on the steering committee for ScAINet, the SeCurity AI Networking conference. She received her Ph.D. in Computer Science from the University of California, Santa Cruz.
Aleatha Parker-Wood
Machine Learning and Algorithmic Privacy Lead, Humu
-
Speaker:Javad Rafiei Asl
Author(s):Javad Rafiei Asl; Sidhant Narula; Mohammad Ghasemigol; Eduardo Blanco ; Daniel Takabi
A Framework for Adaptive Multi-Turn Jailbreak Attacks on Large Language Models
Abstract: This paper introduces HarmNet, a modular framework designed to systematically construct, refine, and execute multi-turn jailbreak queries against LLMs, demonstrating significantly higher attack success rates compared to prior methods.
-
Speaker: Tamás Vörös
Author(s): Tamás Vörös; Adarsh Kyadige
This work proposes LLM salting, a lightweight defense mechanism that rotates the internal refusal direction of LLMs, rendering previously effective jailbreak prompts (like GCG) ineffective without degrading model utility.
-
Speaker: Amelia Kawasaki
Author(s): Amelia Kawasaki; Kasimir Schulz; Leo Ring
Abstract: This paper unveils ShadowLogic, a method for injecting hidden backdoors into white-box LLMs by modifying theircomputational graphs. These backdoors are activated by a secret trigger phrase, allowing the model to generate uncensored responses and exposing a new class of graph-level vulnerabilities.
-
Speaker: Jake Thomas
Author(s): Jake Thomas; Damian Ruck; Gabriel Downer; Sean Craven
Abstract: This research presents Text2VLM, a novel pipeline that adapts text-only datasets into multimodal formats to evaluate the resilience of Visual Language Models (VLMs) against typographic prompt injection attacks. It highlights the increased susceptibility of VLMs when visual inputs are introduced.
-
Speaker: Tim Dudman
Author(s): Tim Dudman
Abstract: This work proposes Topological Extensions for Reinforcement Learning Agents (TERLA) to provide generalizability for cyber defense agents in networks of differing topology and size without the need for retraining. It evaluates performance in realistic simulation environments.
-
Speaker: Ashley Song
Author(s): Ashley Song; Hsin Chen; Shawn Davis; Dhruv Nandakumar
Abstract: This study evaluates Test-Time Compute for improving the accuracy and consistency of real-world cybersecurity agentic systems, specifically a container vulnerability analysis agent and a server alert triage agent.
-
Speaker: Benji Lilley
Author(s): Benji Lilley; Brian Mitchell; Spiros Mancoridis
Abstract: This paper introduces Relational Inference GraphRAG (RIG-RAG), an LLM-assisted pipeline that transforms cloud configuration data into a security-enriched knowledge graph to support natural-language reasoning about deployed infrastructure. This enhances agentic capabilities for cloud security operations.
DAY TWO
Keynote: On Evaluating Adversarial Robustness
Several hundred papers have been written over the last few years proposing defenses to adversarial examples (test-time evasion attacks on machine learning classifiers). In this setting, a defense is a model that is not easily fooled by such adversarial examples. Unfortunately, most proposed defenses to adversarial examples are quickly broken.
This talk examines the ways in which defenses to adversarial examples have been broken in the past, and what lessons we can learn from these breaks. Begin with a discussion of common evaluation pitfalls when performing the initial analysis, it then turns to recommendations for how we can perform more thorough defense evaluations.
Nicholas Carlini
Research Scientist, Google
-
Speaker: Mohammad Saidur Rahman
Author(s): Mohammad Saidur Rahman; Scott Coull; Qi Yu; Matthew Wright
Abstract: This study proposes MADAR, a Continual Learning (CL) framework for malware classification, which addresses catastrophic forgetting by incorporating diversity-aware replay. It demonstrates improved detection accuracy for both Windows and Android malware datasets.
-
Speaker: Supriti Vijay
Author(s): Supriti Vijay; Aman Priyanshu
Abstract: This paper presents a multi-turn retrieval architecture for automating vulnerable code discovery, where models iteratively generate and refine search queries. It introduces a reinforcement learning environment and dataset for training such strategies.
-
Speaker: Christopher Honaker
Author(s): Christopher Honaker
Abstract: This research examines social interactions in open-source code repositories using a biased BERTopic model to identify emerging cybersecurity threats (e.g., the XZ Utils backdoor) by prioritizing negative sentiment and cybersecurity keywords.
-
Speaker: Edward Raff
Author(s): Edward Raff; Karen Kukla; Michel Benaroch; Joseph Comprix
Abstract: This work explores Adversarial Machine Learning (AML) attacks on financial reporting, demonstrating how bad actors can manipulate financial statements to inflate earnings and reduce fraud scores simultaneously, highlighting a critical information security vulnerability in financial systems.
-
Speaker: Sadegh Momeni
Author(s): Sadegh Momeni; Ge Zhang; Birkett Huber ; Hamza Harkous; Sam Lipton; Benoit Seguin; Yanis Pavlidis
Abstract: This paper proposes a two-stage hybrid framework for ML-based threat detection in enterprise security, combining loose YARA rules with an ML classifier and leveraging synthetic data generation (Simula) and active learning to achieve a self-sustained, low-overhead solution for SOCs.
-
Speaker: Steven Leung
Author(s): Steven Leung
Abstract: This study provides the first large-scale empirical evaluation of Risk-Based Authentication (RBA) effectiveness in production two-factor authentication (2FA) systems against real-world opportunistic, targeted, and advanced attacks. It demonstrates how heuristic and anomaly detection methods improve security while maintaining user experience.
-
Speaker: Daniel Jones
Author(s): Daniel Jones; Giorgio Severi; Martin Pouliot; Gary Lopez; Joris de Gruyter; Santiago Zanella-Beguelin; Justin Song; Blake Bullwinkel; Pamela Cortez; Amanda Minnich
-
-
-
Speaker: Gary Lopez Munoz
Author(s): Gary Lopez Munoz; Mauricio Velazco; Manuel MelŽndez
-
-
Speaker: Arjun Krishna
Author(s): Arjun Krishna, Erick Galinkin, Leon Derczynski, Jeffrey Martin
-
Speaker: Subhabrata Majumdar
Author(s): Subhabrata Majumdar; Brian Pendleton; Abhishek Gupta