Tommaso Cerruti

I am an MSc student in Machine Intelligence at ETH Zurich. Previously, I was a Software Engineer Intern at Amazon in the Alexa AI team, where I developed a service to improve the NER capabilities of Alexa AI. I received my BEng in Computer Engineering from Politecnico di Torino, where I was part of the Young Talents Program and was awarded the Agon Scholarship following my national and international achievements in rowing, including an indoor rowing world record.

I am interested in the safety, alignment, and evaluation of autonomous AI agents. In particular, I am interested in three questions:

  1. How can we distinguish agents that are robustly aligned from agents that only appear aligned in the controlled environment of an evaluation?
  2. What new attack surfaces emerge once agents have tools, memory, and long-running state, and how can we diagnose and mitigate attacks such as memory poisoning?
  3. How can malicious or compromised agents manipulate the plans, beliefs, or objectives of benign agents in multi-agent settings?

I have also served as a reviewer for the ICML workshops "Agents in the Wild: Safety, Security, and Beyond", "Compositional Learning: Safety, Interpretability, and Agents", and "Trustworthy AI for Good".

Outside of research, I enjoy sports (rowing, muay thai, weight lifting, and hiking), travelling (the picture on the right was taken in the Sahara Desert, in Morocco), and trying new foods (although nothing beats a carbonara with Coca-Cola).

(Last updated: May 2026)

CV on request via email. Most projects you can find them on GitHub, whereas my publications appear on GScholar.