Xander Davies

Breaking Agent Backbones: Evaluating the Security of Backbone LLMs in AI Agents

Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs

Existing Adversarial LLM Unlearning Evaluations Are Inconclusive

RedCodeAgent: Automatic Red-teaming Agent against Diverse Code Agents

Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples

Fundamental Limitations in Pointwise Defences of LLM Finetuning APIs

AgentHarm: A Benchmark for Measuring Harmfulness of LLM Agents