OpenAI says it signed an AI safety research agreement with Poland’s NASK

OpenAI said on October 23, 2026 that it has signed an agreement with Poland’s NASK National Research Institute to explore joint work on AI safety, security, and technical research. The announcement is OpenAI’s. This desk did not find a separate NASK release to compare it with, so the scope below is what OpenAI published.

Under the agreement, OpenAI says, NASK’s AI Safety Research Center — which it describes as the Polish Artificial Intelligence Safety Institute — and OpenAI will look at how to evaluate more capable systems, find vulnerabilities, and strengthen safeguards. The post does not attach a budget, a timetable, a list of models, or a commitment to publish results.

What the two sides said in the post

NASK director Radosław Nielek, quoted by OpenAI, said the institute’s safety work includes safeguards against emerging risks, detecting synthetic content, and studying malicious uses and multilingual issues. He said AI safety “cannot be achieved in isolation,” and that public safety institutes and frontier developers have to cooperate if the systems are going to be reliable.

Tom Duff Gordon, OpenAI’s vice president and head of EMEA policy, said many of the company’s researchers have come from Poland, and that the agreement is a way to test OpenAI’s assumptions with outside experts. He called it a door into “the international effort to make AI safer.”

OpenAI places the deal next to work it says it already does with the US Center for Advancing Standards and Innovation for Super Intelligence and the UK AI Security Institute. It argues for shared ways to evaluate capabilities, assess risks, test safeguards, and report incidents, so governments can compare evidence across borders. That is the company’s case for the partnership. The post does not say Poland has adopted a new statute.

The July arrangement OpenAI cites

The new agreement, OpenAI says, follows access that CERT Polska, which operates inside NASK, received in July through a program OpenAI calls Daybreak. OpenAI says the team used its models to help find six vulnerabilities in MikroTik RouterOS, including flaws that could be combined to take control of affected devices without authentication. OpenAI says the vendor shipped fixes and CERT Polska confirmed those fixes blocked the attacks it had observed.

This article is not a vulnerability write-up. OpenAI’s post does not name the flaw identifiers, the versions, or the exploit steps, and those details are not repeated here. The safety agreement announced on October 23 is a research collaboration, as OpenAI describes it, not a claim that either organization has certified any model as safe.

This report is not legal, financial, or health advice. See the disclaimer.