Lakera's research team is on the mission to secure the Internet of Agents. We uncover fundamental AI vulnerabilities, push the limits of adversarial AI, and develop defenses that reshape how AI systems withstand attacks. Our work combines cutting-edge research with real-world impact, setting new standards for securing autonomous systems.
Latest Research Updates
This section will be regularly updated with insights from our red teaming efforts, including new findings, methodologies, interactive demos, and potential attack vectors that we uncover.
Gandalf: Adaptive Defenses for Large Language Models
This research introduces D-SEC, a threat model that separates attackers from legitimate users and captures dynamic, multi-step interactions. Using Gandalf—a crowd-sourced red-teaming platform—we analyze 279k real-world attacks and show how some defenses degrade usability. We highlight effective strategies like adaptive defenses and defense-in-depth.
Join Us in Securing the Future of AI
We invite researchers, developers, and security professionals to collaborate with us. Whether you’re interested in contributing to our projects, testing new defense strategies, or exploring novel AI security concepts, we welcome you to join us.
Contact Us