======
Our work was the first to discover and coin indirect prompt injection attacks, which is currently a major vulnerability of LLMs, and propose watermarking generative AI models. Our work also won a best paper award at ACL 2025 (awarded to 4 papers out of more than 3000 accepted ones).
Open Positions
============
I will have multiple positions and hiring on all levels. I am hiring PhD students, post-docs, research interns (predoctoral), or visiting PhD students. Please fill out this form if you are interested to work with me!
If you have a Master's degree and would like to apply for a PhD, you can apply directly via this form.
I will also admit PhD students via the following channels:
Broad Research Vision:
==================
Our group will work on ensuring that the next-wave of advanced super-intelligent AI (or AGI) is aligned with human and societal values. Particularly, we are interested in advanced autonomous AI agents, rigorously evaluating and minimizing their safety implications, and how to use them for good to improve our society. Our group will be highly interdisciplinary: from computer science, social science, economics, and cognitive science fields. We will also maintain strong connections with industry and governmental AI safety research institutes.
Research Style:
=============
To achieve this vision, we need to think ahead to proactively ensure the safety of advanced AI while being grounded with today's needs. We are interested in foundational and impactful research by uncovering fundamental limitations, offering a new perspective, and improving our understanding.
Specific Research Directions:
=======================
Any direction that serves this research vision is relevant to our group. We value curiosity and flexibility. Specific examples are indicated below:
- Contextual agents (that respects privacy and safety norms and reason about their evolving context)
- Red-team and securing real-world systems
- Agents with long-term memory
- Self-evolving and open-ended agents
- Cooperative agents
- Interpretability and white-box control methods
- Building evaluations for new capabilities
- Deceptive alignment and situational awareness
- Gradual loss of control due to long-term interaction with agents
- Opportunities: AI for mediation and finding common grounds
- Opportunities: AI for scientific discovery
Some of my papers that are relevant:
Multi-agent cooperation,
Test awareness,
Open-ended safety,
Contextual integrity reasoning
Requirements:
============
For PhD students, I am looking for candidates (with a Master's degree) who are passionate and willing to learn about AI Safety problems and familiar with current research problems via seminars, MSc thesis, AI safety camp, conferences, course projects, etc. Publication record is nice to have but not required; passion for impactful research is particularly appreciated.
Highlights:
=========
- Tübingen has an amazing environment for AI research (and also AI safety) with so many great researchers!! Collaborations are very welcomed!
- I have ongoing collaborations with industry teams which I am planning to maintain and strengthen
- No knowledge of German language is needed
I read all emails and applications, but responses may take time. I highly appreciate your time and thank you for your interest! I will reach out if there is a fit. Due to the large volume of applications, I may not be able to respond individually. Thank you for understanding.
