OpenAI and Poland's NASK National Research Institute have signed an agreement to explore collaboration on AI safety, security and technical research, OpenAI announced on October 23, 2026. The announcement places the work within a wider effort to evaluate increasingly capable AI systems and strengthen their safeguards.
What the agreement covers
According to OpenAI, NASK's AI Safety Research Center, which serves as the Polish Artificial Intelligence Safety Institute, will work with the company on AI safety research. The stated areas include evaluating advanced systems, identifying potential vulnerabilities and improving safeguards. The announcement does not give a project timetable, budget or list of published joint deliverables.
NASK director Radoslaw Nielek describes the center's work as including safeguards against emerging risks, detection of synthetic AI-generated content, malicious uses of that content and multilingual challenges. Those areas explain why national research institutions can bring different testing priorities from a model developer's own internal evaluations.
The Polish cybersecurity connection
The announcement also describes an earlier connection with CERT Polska, which operates within NASK. OpenAI says the team gained access to its cyber-capable models through Daybreak in July and used them to help discover six vulnerabilities in MikroTik RouterOS.
According to the same account, some of those flaws could be combined to take control of affected devices without authentication. OpenAI says the vendor released fixes and CERT Polska confirmed that the fixes prevented the attacks it observed. That is background to the new agreement, not evidence that the agreement itself has already produced those findings.
Why shared evaluations matter
OpenAI frames the agreement as part of international work on technical standards for frontier AI. Its announcement points to common approaches for evaluating capabilities, assessing risks, testing safeguards and reporting incidents. Shared methods could help institutions compare evidence across borders, but this announcement does not itself introduce a binding international standard.
For companies building with AI, the immediate news is a new research relationship rather than a change to API access or product terms. The practical questions to watch are which evaluations the partners publish, what they reveal about multilingual risks and whether findings lead to specific improvements in deployed systems. Until those outputs appear, the agreement should be read as a commitment to explore joint work, not a completed safety assessment.