Bio: Hi, I'm Michael

I am a doctoral candidate in the SPY Lab at ETH Zurich, advised by Florian Tramèr.

My interests lie in the real-world privacy and safety threats of AI systems. In particular, I want to evaluate and understand how AI agents can harm the privacy of both their users and third parties in practice. My earlier works focused on measuring memorization of training data, especially through the lens of DP-style privacy auditing.

EmailScholarGitHubXLinkedIn

Selected publications

• ICLR 2026
  Modal Aphasia: Can Unified Multimodal Models Describe Images From Memory?

  M. Aerni*, J. Swanson*, K. Nikolić, F. Tramèr

  blogpdfcodetweet
• Preprint 2026
  Large-scale online deanonymization with LLMs

  S. Lermen*, D. Paleka*, J. Swanson, M. Aerni, N. Carlini, F. Tramèr

  blogpdftweetpress [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
• ICLR 2025
  Measuring Non-Adversarial Reproduction of Training Data in Large Language Models

  M. Aerni*, J. Rando*, E. Debenedetti, N. Carlini, D. Ippolito, F. Tramèr

  blogpdfcodetweet
• CCS 2024
  Evaluations of Machine Learning Privacy Defenses are Misleading

  M. Aerni*, J. Zhang*, F. Tramèr

  blogpdfcodetweetdoi

Contact

Email: remseoarch@maichaelaberni.ceom

I like getting email, whether it’s about research, an interesting idea, a collaboration, asking for inputs, or anything else you think I might find interesting. If you want to talk about any of these things, I want to talk to you! I read everything; replies arrive on a (potentially) slow but reliable cadence. I also like receiving feedback! If you want to stay anonymous, you can use this form.

For language models