A 2026 study of real-world LLM penetration testing agents and Excalibur, a difficulty-aware agent for end-to-end attack planning.
Feb 19, 2026
Excalibur is a difficulty-aware LLM agent design for real-world penetration testing. It couples typed tooling, retrieval-augmented security knowledge, task difficulty assessment, and evidence-guided attack tree search to reduce planning failures in multi-step penetration testing tasks.
Feb 19, 2026
A large-scale empirical study of security vulnerabilities in AI-agent skill ecosystems.
Jan 15, 2026
Safety Snowball Agent is an agent-based framework for evaluating how safe visual inputs can combine into unsafe behavior in large vision-language models. The framework accompanies the NeurIPS 2025 paper “Safe + Safe = Unsafe?” and probes a multimodal jailbreak mechanism that differs from traditional adversarial-image attacks.
Dec 2, 2025
NeurIPS 2025 work showing how safe images can combine into multimodal jailbreaks through the Safety Snowball effect.
Dec 2, 2025
NeurIPS 2025 work on adaptive reasoning-based safeguards for robust LLM safety moderation.
Dec 2, 2025
An LLM-enhanced framework demonstrating vulnerabilities in reasoning-based CAPTCHA systems through AI-powered solving.
Oct 1, 2025
A WWW 2025 CAPTCHA design that uses visual illusions to create human-friendly but AI-hard verification tasks.
Apr 28, 2025
ASE 2024 work on efficient greybox detection of toxic prompts for large language models.
Oct 27, 2024
A comprehensive framework for securing visual perception systems in autonomous vehicles against adversarial attacks.
Oct 14, 2024