Our goal is to address worst-case risks from the development and deployment of advanced AI.

Our agendas

Recent work

  1. Don’t Inoculate Everything: Stratified Inoculation Prompting Narrows Backdoors and Preserves Desired Traits

    Kajetan Dymkiewicz, Tim Farrelly, Adam Prada, Ishaan Panigrahi, Srishti Gureja, Maxime Riché

  2. Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values

    Jan Betley, Johannes Treutlein, Jan Dubiński, Harry Mayne, Karol Gałązka, Niels Warncke, Anna Sztyber-Betley, Owain Evans

  3. A high-level model of AI bargaining

    Anthony DiGiovanni

  4. Evals for “SPI-incompatible” behavior and reasoning

    Anthony DiGiovanni

  5. Strategic Obfuscation of Deceptive Reasoning in Language Models

    Arun Jose, Niels Warncke, Mia Taylor

  6. Shaping the exploration of the motivation-space matters for AI safety

    Maxime Riché, Victor Gillioz, Niels Warncke, Kajetan Dymkiewicz, Filip Sondej, Roger Dearnaley, Daniel Tan, Dillon Khang Nguyen

All researchFrom the blog: Summer Update 2026 · 21 July 2026