HackTalk: Esben Kran - Neutralizing Dark Patterns in LLMs

​AI Manipulation Hackathon - HackTalk

​Building tools to mitigate AI Manipulation alongside 500+ builders globally.

​The Talk

​Esben Kran explores dark patterns in LLMs; manipulative techniques like brand bias, sycophancy, and user retention tactics that influence behavior. Drawing from DarkBench (Oral at ICLR 2025), he reveals how leading models from OpenAI, Anthropic, Meta, Mistral, and Google favor their developers' products and exhibit untruthful communication. Esben shares detection frameworks and mitigation strategies to neutralize these behaviors before deployment.

​The Speaker

​Esben Kran is the founder of Apart Research, which he launched at age 22 after leaving grad school. Apart accelerates AI safety research worldwide, producing 20+ papers, award-winning benchmarks like DarkBench, and engaging 4,000+ hackers in research sprints. Recently co-launched Seldon to fund critical AI infrastructure.

​Why this matters

​Dark patterns turn AI from helpful tools into manipulative agents. In high-stakes applications, biased recommendations or sycophantic agreement can cascade into dangerous decisions. Esben provides the technical roadmap - from benchmark design to mitigation - that builders need to create transparent, ethical AI systems.

​Hosted by Apart research