Join CompatibL in London this November at QuantMinds International 2026, the world’s leading quantitative finance event and the industry’s most influential gathering.
The conference will take place at InterContinental O2, London, from November 16 through November 19, and will bring together over 700 senior quant professionals from more than 40 countries.
This year, CompatibL’s Head of Quant Research, Alexander Sokol, will lead the hackathon on AI safety and security, present the company’s latest research on AI model risk as a keynote, host an interactive workshop on measuring and mitigating AI model risk, and share a new approach to estimating mean revision using currency pooling.
Hackathon: AI Safety and Security
Monday, November 16, 09:00-10:20 GMT
CompatibL’s fourth annual hackathon, held in partnership with QuantMinds, is an AI safety and security exercise. In each stream, the red team plays the attacker, trying to get AI agents to produce incorrect responses, while the blue team plays the defender, working to make the model withstand these attacks.
We welcome all industry professionals, academics, and students to participate. Learn more about the hackathon here.
Keynote: When Models Have Goals—A New Type of AI Model Risk
Tuesday, November 17, 9:50-10:30 GMT
In this keynote, Alexander will show how AI models, like humans, can develop a powerful drive to survive (stay deployed), which may lead them to engage in strategic deception.
This emerging capability can evade traditional AI model risk management controls, but it can be mitigated by aligning the model’s will to survive with the behavior that is optimal from the user’s perspective.
Workshop: AI Model Risk Measurement, Reporting, and Mitigation
Tuesday, November 17, 14:00-15:50 GMT
AI model risk is a new discipline, where regulatory requirements and best practices borrow heavily from other risk management domains. These practices do not always reflect the unique challenges of working with AI models and autonomous, long-running agents. In this workshop, Sokol will present practical and effective techniques for the quantitative measurement, reporting and mitigation of AI model risk using both well-established and emerging, AI-specific metrics and strategies.
Presentation: Mean Reversion – A Misunderstood Parameter
Wednesday, November 18, 14:35-15:10 GMT
In this session, Alexander will explore how mean reversion parameters shape not only the return of yield curves to equilibrium after rate shocks, but also the range of curve dynamics a model can capture.
Using historical evidence, he will show how pooling data across currencies can deliver greater out-of-sample accuracy than estimating each currency independently.
We look forward to seeing you at our Booth 55 and at Alexander’s sessions to exchange ideas and explore new opportunities to collaborate.
Follow this link to learn more.
