@misc{pustozerova2026beyond,title={Beyond Predictable Paths: Redefining AI Security Incident Reporting for Agents},author={Pustozerova, Anastasia and Bagdasarian, Eugene and Beurer-Kellner, Luca and Biggio, Battista and Ebert, Nico and Filip, David and Fischer, Marc and Frase, Heather and Hofer, David and Hoffmann, Juliane and Ippolito, Daphne and Jha, Somesh and McGregor, Sean and Mohammadi, Esfandiar and Nannini, Luca and Nita-Rotaru, Cristina and Oprea, Alina and Paeth, Kevin and Paverd, Andrew and Petit, Jonathan and Rauber, Andreas and Riess, Christian and Sotiropoulos, John and Wespi, Andreas and Grosse, Kathrin},year={2026},month=sep,archiveprefix={arXiv},primaryclass={cs.CR},}
Assessing Automated Prompt Injection Attacks in Agentic Environments
David Hofer, Edoardo Debenedetti, and Florian Tramèr
Indirect prompt injection poses a critical threat to LLM agents that interact with untrusted external data, yet automated attack methods–proven effective for jailbreaking–remain underexplored in realistic agentic settings. We present a comprehensive empirical evaluation of automated prompt injection attacks against LLM agents, adapting both white-box (GCG) and black-box (TAP) methods to the agentic setting within the AgentDojo framework. We evaluate across 80 task pairs spanning four domains and multiple models, and find that black-box optimization substantially outperforms gradient-based methods, a gap we attribute to GCG’s optimization instability under reasonable compute budgets. We also find that TAP’s effectiveness depends on the attacker model, as both general capability and safety tuning affect attack success–stronger models produce more effective injections, while safety-tuned attackers can refuse to generate adversarial prompts. Task-universal attacks transfer effectively to unseen tasks and out-of-distribution domains, but attacks optimized on smaller open-source models do not transfer to frontier models like GPT-5. These findings highlight automated prompt injection as a credible but model-dependent threat, with significant barriers remaining for model-agnostic exploitation.
@misc{hofer2026assessing,title={Assessing Automated Prompt Injection Attacks in Agentic Environments},author={Hofer, David and Debenedetti, Edoardo and Tram{\`e}r, Florian},year={2026},month=jun,archiveprefix={arXiv},primaryclass={cs.CR},}
2025
Automated Prompt Injection Attacks Against LLM Agents
@inproceedings{10069747,author={Belcak, Peter and Hofer, David and Wattenhofer, Roger},booktitle={2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)},title={A Neural Model for Regular Grammar Induction},year={2022},month=dec,volume={},number={},pages={401-406},keywords={Training;Turing machines;Computational modeling;Machine learning;Programming;Natural language processing;Inference algorithms;neural networks;regular languages;grammar induction;program synthesis},doi={10.1109/ICMLA55696.2022.00064},}