MD-ML: Super Fast Privacy-Preserving Machine Learning for Malicious Security with Dishonest Majority
Explore groundbreaking advances in privacy-preserving machine learning with MD-ML framework, offering enhanced security and efficiency for multi-party computation in neural network inference and training.
DNN-GP: Diagnosing and Mitigating Model's Faults Using Latent Concepts
Explore an innovative fault diagnosis tool for Deep Neural Networks that interprets latent concepts to detect and mitigate various model vulnerabilities, from adversarial attacks to concept drift.
Lotto: Secure Participant Selection against Adversarial Servers in Federated Learning
Explore secure participant selection in Federated Learning through Lotto, a system designed to protect against adversarial servers while maintaining efficient performance and privacy guarantees.
Defending AI on Kubernetes - A Red Team Security Playbook
Explore offensive security strategies and defense mechanisms for AI systems on Kubernetes, focusing on vulnerability assessment and protective measures from an attacker's viewpoint.
AI Under the Hood: Unmasking Hidden Threats
Explore a comprehensive analysis of AI threat modeling frameworks, examining safety protocols and risk assessment methods for AI/ML applications while learning to identify and address potential vulnerabilities.
Safeguarding Machine Learning Systems - A Comprehensive Security Plan
Discover a holistic security framework for ML systems, covering data protection, model fortification, platform security, compliance measures, and human-centric safeguards for robust AI deployment.
Instruction Backdoor Attacks Against Customized Large Language Models
Explore groundbreaking research on instruction backdoor attacks targeting customized LLMs, examining vulnerabilities in GPT applications and proposing effective defense strategies against security threats.
Formalizing and Benchmarking Prompt Injection Attacks and Defenses
Explore a systematic framework for understanding and evaluating prompt injection attacks on LLM applications, including analysis of 5 attack types and 10 defense mechanisms across multiple models.
Tossing in the Dark: Practical Bit-Flipping on Gray-box Deep Neural Networks for Runtime Trojan Injection
Explore practical bit-flipping techniques for runtime Trojan injection in deep neural networks, focusing on gray-box scenarios and innovative approaches to bypass memory protection systems.
PhishDecloaker: Detecting CAPTCHA-cloaked Phishing Websites via Hybrid Vision-based Interactive Models
Discover how PhishDecloaker uses AI and computer vision to detect CAPTCHA-protected phishing websites, improving detection rates from 0% to 74.25% and enhancing cybersecurity defenses.
AI Supply Chain Security - Inventing Backward and Forward
Explore emerging AI supply chain security threats and defensive strategies, focusing on CI/CD pipelines and creative approaches to protect AI-enabled features in modern software systems.
ShadowRay: First Known Attack Campaign Targeting AI Workloads Actively Exploited in the Wild
Discover how attackers exploit AI infrastructure through the ShadowRay campaign, examining real-world cases, attack techniques, and critical security lessons for protecting AI workloads.