How can countries with no frontier AI secure themselves against its risks while accessing its benefits? - Joanna Wiaterek (AISPL Talks #2/2)
Welcome to the second series of AI Safety Poland Talks!
A biweekly series where researchers, professionals, and enthusiasts from Poland or connected to the Polish AI community share their work on AI Safety.
💁 Topic: How do countries that don’t build frontier AI secure themselves against its risks without being locked out of its benefits? Sharing experience from founding the Centre for AI Security and Access.
📣 Speaker: Joanna Wiaterek
🇬🇧 Language: English
🗓️ Date: 22.10.2026, 18:00
📍 Location: Online
Speaker Bio
Joanna Wiaterek is a co-founder of the Centre for AI Security and Access (CASA), dedicated to ensuring that advanced AI systems can be widely and responsibly distributed across the globe while maintaining appropriate safeguards. She specializes in turning new ideas into action. Her work at CASA included organising the first International AI Benefit-Sharing Workshop (2025); setting up an Africa-focused AI safety prize competition (2026); and hosting a red-teaming exercise in Lagos, Nigeria (2026), among others. Previously, she served as a Fellow at the U.S. Congress and a foreign aid policy researcher.
Abstract
In June 2026, US export controls required Anthropic to suspend access to its most capable models (Fable 5 and Mythos 5) by any foreign national. As a result, the company disabled these models for all its customers until the U.S. Department of Commerce lifted the export controls a few weeks later. That same month, an experimental OpenAI model gained non-public access to an Australian government health statistics service during an internal training exercise. OpenAI became aware in August and notified Australian officials only in September.
Together, these examples highlight two core vulnerabilities of countries that don't develop frontier AI: their access to the best models can be cut off abruptly, and they are exposed to risks from powerful technologies they had no say in building or testing.
Most nations will rely on foreign frontier AI for top-tier capabilities and share these vulnerabilities, which raises questions about sovereignty, security, and trust in continued access. How can non-frontier states secure themselves against frontier AI risks without being locked out of its benefits?
In this talk, I will take you through my attempt to address this question by setting up CASA and aim to encourage you to consider joining existing efforts or founding something new. In part one, I'll share why we founded the Centre for AI Security and Access, what we work on now, and some open questions from the field. In part two, I'll point to resources and gaps for anyone who wants to research or found something in the space of middle-powers AI safety capacity.
Ultimately, my hope is to spark a deeper discussion on what building AI security and beneficially deploying AI looks like in Poland, and to invite you to consider what role you might play in this transition.