Get started
Author

Xander Davies

Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
2023 · Casper, Stephen, Davies, Xander, et al. · arXiv (Cornell University) · 879
PDF
AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
2024 · Andriushchenko, Maksym, Souly, Alexandra, et al. · arXiv (Cornell University) · 616
PDF
Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples
2025 · Souly, Alexandra, Rando, Javier, et al. · arXiv (Cornell University) · 144
PDF
Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs
2025 · O'Brien, Kyle, Casper, Stephen, et al. · arXiv (Cornell University) · 49
PDF
Unifying Grokking and Double Descent
2023 · Davies, Xander, Langosco, Lauro, et al. · arXiv (Cornell University) · 47
PDF
How Vulnerable Are AI Agents to Indirect Prompt Injections? Insights from a Large-Scale Public Competition
2026 · Dziemian, Mateusz, Lin, Maxwell, et al. · arXiv (Cornell University) · 31
PDF
Circuit Breaking: Removing Model Behaviors with Targeted Ablation
2023 · Li, Maximilian, Davies, Xander, et al. · arXiv (Cornell University) · 31
PDF
Discovering Variable Binding Circuitry with Desiderata
2023 · Davies, Xander, Nadeau, Max, et al. · arXiv (Cornell University) · 27
PDF
Sparse Distributed Memory is a Continual Learner
2023 · Bricken, Trenton, Davies, Xander, et al. · arXiv (Cornell University) · 19
PDF
STACK: Adversarial Attacks on LLM Safeguard Pipelines
2025 · McKenzie, Ian R., Hollinsworth, Oskar J., et al. · arXiv (Cornell University) · 13
PDF
Fundamental Limitations in Pointwise Defences of LLM Finetuning APIs
2025 · Xander Davies, Eric Winsor, et al. · neural information processing systems · 13
PDF
Evaluating whether AI models would sabotage AI safety research
2026 · Kirk, Robert, Souly, Alexandra, et al. · arXiv (Cornell University) · 12
PDF
Boundary Point Jailbreaking of Black-Box LLMs
2026 · Davies, Xander, Giglemiani, Giorgi, et al. · arXiv (Cornell University) · 12
PDF
RedCodeAgent: Automatic Red-teaming Agent against Diverse Code Agents
2025 · Guo, Chengquan, Xie, Chulin, et al. · arXiv (Cornell University) · 12
PDF
Existing Large Language Model Unlearning Evaluations Are Inconclusive
2025 · Feng, Zhili, Xu, Yixuan Even, et al. · arXiv (Cornell University) · 10
PDF
Gaming AI-Assisted Peer Reviews Poses New Risks to the Scientific Community
2026 · Li, Lin, Zhang, Qi, et al. · arXiv (Cornell University) · 6
PDF
UK AISI Alignment Evaluation Case-Study
2026 · Souly, Alexandra, Kirk, Robert, et al. · arXiv (Cornell University) · 6
PDF
SECODEPLT: A Unified Benchmark for Evaluating the Security Risks and Capabilities of Code GenAI
2025 · Yuzhou Nie, Zhun Wang, et al. · neural information processing systems · 2
STACK: Adversarial Attacks on LLM Safeguard Pipelines
2026 · Ian R. McKenzie, Oskar John Hollinsworth, et al. · Proceedings of the AAAI Conference on Artificial Intelligence · 1
Security Challenges in AI Agent Deployment: Insights from a Large Scale Public Competition
2025 · Andy Zou, Maxwell Lin, et al. · neural information processing systems · 1
Evaluating Whether GPT-6 Astra Performs Unsanctioned Supply-Chain Attacks
2026 · Souly, Alexandra, Fronsdal, Kai, et al. · arXiv (Cornell University)
PDF
1174 - STACK: Adversarial Attacks on LLM Safeguard Pipelines
2026 · Association for Artificial Intelligence 2026, Casper, Stephen, et al. · Underline Science Inc.
STACK: Adversarial Attacks on LLM Safeguard Pipelines
2026 · Association for Artificial Intelligence 2026, Casper, Stephen, et al. · Underline Science Inc.
Understanding and Enhancing Backdoor Persistency in LLM Agent Post-Training
2026 · Qiusi Zhan, Nian Lyu, et al. · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)
PDF
Works count: 28
Citation count: 1,935
H-index: 12
I10-index: 16
Year
Topic
Type