W4AIS | Women for AI Safety
W4AIS | Women for AI Safety
Date: Wednesday 29th July
Time: 19:00 to 20:00 CEST
Location: Zoom (link shared upon registration)
Organised by: BlueDot Impact
Facilitated by: Osmani Redondo and Mmachukwu Osisioma
Title:
The model passed, but should we believe it?
Subtitle:
Contamination, memorization, and trustworthy safety evals
About the Event
Most of the people shaping AI safety and governance today come from a narrow set of backgrounds and perspectives. That is not just a question of representation, it is a question of what gets protected and what gets missed.
W4AIS exists to widen that group, now, while the field is still being defined.
Whether you work in AI safety, alignment, governance or policy, or you are simply paying attention to where this is going, this is your space.
We are giving foundation models exams whose answer keys are in the training data — and selecting the AI infrastructure of society based on the results.
This session covers contamination, memorization, mitigation strategies, and what frontier labs are actually doing about it. You'll leave with a better question to ask the next time a model "passes."
Agenda (1 hour)
No preparation needed. Just bring curiosity and enthusiasm. If you'd like to share something, you can propose a talk for future sessions.
Talk + Questions
Closing and next steps (10 mins): announcements and how to stay connected
Session materials will be available exclusively to registered participants via W4AIS. The session will be recorded.
W4AIS aims to gather regularly, as a safe, accessible and inclusive space to connect, share ideas and learn together. Everyone interested in AI Safety is welcome.
Speaker
Johanna built and led the model-evaluation function at a European AI lab from zero, owning release gating and safety benchmarking across LLMs and quantum-compressed models. She now engineers AI governance into production for regulated enterprises, translating EU AI Act and ISO 42001 obligations into technical controls. She holds two M.Sc. degrees in Artificial Intelligence and a B.Sc. in Computer Science, and is a PhD candidate building a neuro-symbolic agent for variant prioritization in rare diseases — along with the contamination-controlled benchmarks needed to trust it.
Follow us on LinkedIn.
Thank you, and see you soon!