OpenAI and Poland’s NASK Agree to Explore AI Safety Research

OpenAI and Poland’s NASK National Research Institute announced an agreement on October 23 to explore collaboration on AI safety, security and technical research. The announcement adds a national public research institution to the work of evaluating increasingly capable systems and strengthening their safeguards.

In its statement, OpenAI says the collaboration will involve NASK’s AI Safety Research Center, which serves as Poland’s Artificial Intelligence Safety Institute. The intended work includes assessing systems, finding vulnerabilities and improving protections. The statement does not specify a funding commitment or a delivery timetable.

NASK’s own announcement emphasizes research into synthetic content, harmful uses and risks that vary across languages. Those are useful distinctions for a field in which an evaluation in one language may leave important questions about another unanswered.

The next evidence should be research output

The agreement is a commitment to explore work together. It does not certify an OpenAI product as safe or establish that a new safeguard has already passed an independent assessment. For technology buyers, that boundary is worth keeping clear when a research partnership appears in a supplier presentation.

The practical value could come from a more precise account of failure: what a system did, under which conditions, and whether a proposed fix still worked when the same test was repeated. A useful public result would let readers separate improved behavior from a narrower test or a changed scoring method.

Our reading is that language coverage could be one of the collaboration’s most consequential areas. Businesses deploying an assistant across countries need to know whether its handling of sensitive requests remains consistent. Simply translating an English test set would not, by itself, establish that local misuse patterns have been covered.

What to watch as the partnership develops

The next announcements should be assessed for concrete research scope, methods and publication plans. Who chooses the tests? Can researchers report negative findings? Which results can outside teams reproduce? These questions concern the eventual work; the agreement alone does not answer them.

Public institutions can bring knowledge of the services, languages and communities in which AI operates. Developers can bring access to systems that researchers need to examine. The useful outcome is evidence that either side could challenge, with limitations visible to customers and other researchers.

For now, the confirmed development is an agreement between OpenAI and NASK. The substantive test comes when that relationship produces evaluations, fixes or shared methods that others can inspect.

Image: OpenAI