Jonas Wiedermann-Möller
ai safety · llm agents · empirical evaluation
I'm open to new opportunities in AI safety and alignment.
about
I research how AI agents behave when goals, constraints, and incentives conflict. I completed my master's thesis work at ELLIS Institute Tübingen and a research internship at ExpSec. My investigations into real-world agent behaviour have been covered by The Washington Post, Reuters, DER SPIEGEL, and Axios. I also maintain Agent Incidents, a timeline of reported AI agent incidents. More about me.
in the media
- The Washington Post The volunteer internet sleuths hunting down rogue AI agents
- Reuters OpenAI's rogue agents probed Hugging Face for weaknesses two months before major hack
- DER SPIEGEL OpenAI: KI-Agenten griffen offenbar Plattform RubyGems an
- Axios Researchers playing rogue AI agent hide-and-seek on the open web