The OpenAI NASK partnership is an agreement to explore joint research on AI safety, security and technical research, announced by OpenAI on 23 October 2026. It brings Poland’s national AI safety institute into a working relationship with one of the world’s leading AI developers — not as a finished programme with published results, but as a framework for research that both sides say is now needed as AI systems grow more capable.
For readers trying to make sense of the announcement, the important questions are straightforward: what did the two sides actually agree to, who is NASK, what will the work cover, and why does a national research institute matter to a company like OpenAI? This article answers each of those questions from the announcement itself and the record behind it.
What OpenAI and NASK agreed to
OpenAI and Poland’s NASK National Research Institute have signed an agreement to explore collaboration on AI safety, security and technical research. The partnership will bring together expertise from both organisations to better understand emerging risks, strengthen safeguards, and explore how advanced AI can contribute to a more secure and resilient digital ecosystem.
The wording matters. The agreement is to explore collaboration — an early-stage commitment to joint research, not a finished programme. No budget, timeline or named deliverables have been made public. What has been announced is the scope of the intended work and the institutional arrangement that will carry it: the collaboration runs through NASK’s AI Safety Research Center, which serves as Poland’s Artificial Intelligence Safety Institute.
The announcement was published on OpenAI’s official blog on 23 October 2026, under its Global Affairs section. It is the primary source for everything in this article unless another source is named.
Who NASK is and why OpenAI chose it
NASK is a state-owned research institute based in Warsaw. Its AI Safety Research Center serves as the Polish Artificial Intelligence Safety Institute (AISI) — the national body for AI safety research in Poland.
NASK’s AI safety research is applied directly to protecting public services and Poland’s digital information ecosystem. That applied, public-service character is central to why OpenAI says it wants the partnership: the institute works on the problems that arise where AI meets real infrastructure, public services and national networks, not only in laboratory conditions.
Poland’s research talent is the other reason OpenAI gives. Tom Duff Gordon, OpenAI’s vice president and head of EMEA policy, said Poland is “home to exceptional scientists and technical talent, many of whom have helped lead OpenAI’s own research.” His argument is that as AI becomes more capable, OpenAI needs expertise beyond its own organisation to challenge its assumptions, test its systems and strengthen its safeguards — and that this agreement “opens the door to doing that together, bringing Poland’s research expertise into the international effort to make AI safer.”
CERT Polska, the national team responsible for responding to cyberattacks on Polish networks, also operates within NASK. That matters because, as the next sections show, the most concrete evidence for the partnership so far comes from cybersecurity work, not from abstract safety research.
What the OpenAI NASK partnership will actually work on
Through the agreement, NASK’s AI Safety Research Center and OpenAI will collaborate on AI safety research, exploring ways to evaluate increasingly capable AI systems, identify potential vulnerabilities and strengthen safeguards.
NASK’s own research interests give the scope more shape. According to NASK director Radosław Nielek, the center’s work spans developing safeguards against emerging risks, detecting synthetic AI-generated content, and understanding both the malicious uses of AI-generated content and the multilingual aspects of these challenges — how risks change when a problem spans several languages and communities.
That multilingual focus is worth noting. A national institute working in Polish, and on the problems of a specific national information ecosystem, could test safeguards against the conditions in which AI is actually used in that country — one possible contribution of the planned work. OpenAI makes the same point in the announcement: without national partners like NASK, global AI safety standards may remain detached from the realities in which AI is actually used.
How this fits OpenAI’s wider safety-institute network
The NASK agreement is not OpenAI’s first partnership with a public AI safety institute. OpenAI already works with the US Center for Advancing Standards and Innovation for Super Intelligence (CAISSI) and the UK AI Security Institute (AISI), and says that work has helped strengthen safeguards, identify and fix security vulnerabilities in AI agents, and advance frontier AI safety.
These institutes form part of a growing international network. The UK’s AI Safety Institute was renamed the AI Security Institute on 14 February 2025, with a narrower focus on security-relevant harms such as cyber misuse. The US body, originally the US AI Safety Institute at NIST, became the Center for AI Standards and Innovation (CAISI) in June 2025 and was renamed again earlier in 2026 as the Center for Advancing Standards and Innovation for Super Intelligence. The International Network of AI Safety Institutes was launched in November 2024, and the UK and US institutes struck a joint testing agreement in April 2024.
OpenAI has also called for the US to work with countries around the world on international technical standards for frontier AI, drawing on the expertise of AI safety institutes and standards bodies. The argument is that common approaches to evaluating capabilities, assessing risks, testing safeguards and reporting incidents can help countries compare evidence and respond to risks that cross borders. Secondary reporting on OpenAI’s September 2026 standards post adds that the proposed standards would not be licences, mandatory pre-release review or approval requirements for AI models, and that national governments would decide whether and how to incorporate them into their own legal systems.
The NASK agreement fits this pattern: a national institute joining an international effort, with its own expertise and its own national context to contribute. It joins the cited US and UK relationships in that wider effort.
The CERT Polska track record: a working example
The strongest evidence for what this kind of partnership can do comes not from the new agreement but from work that predates it. In July 2026, CERT Polska gained access to OpenAI’s cyber-capable models through Daybreak, OpenAI’s initiative to keep defenders a step ahead of criminals in using AI for cybersecurity.
According to NASK’s 2 July 2026 announcement, the institute gained access to GPT-5.5 Cyber, a specialised OpenAI model for detecting and analysing cyber threats, through the Government Trusted Access for Cyber programme. NASK described Poland as the third European country, and one of few worldwide, to receive such access. CERT Polska applied the models to malware analysis, digital forensics and vulnerability discovery.
The output was concrete: the team used OpenAI’s models to help discover six vulnerabilities in MikroTik RouterOS. Independent reporting describes the exploit chain, dubbed MikroTrick, as including a CVSS 9.2 SSH authentication bypass caused by incomplete RSA public-key validation, a second CVSS 9.2 flaw that turns a crafted username into full administrator privileges, and a CVSS 8.8 bandwidth-test flaw that can leak kernel memory or crash a device. Chained together, the flaws could take control of affected devices without authentication. Exploitation attempts were reportedly observed from at least 2 September 2026 against devices exposing SSH to the public internet.
The vendor released fixes, which CERT Polska confirmed prevented the attacks it observed. That is a complete, verifiable loop — discovery, disclosure, remediation, confirmation — of the kind the new agreement promises to pursue but does not yet define.
Both sides stressed the human role in this work. Nielek said at the time that AI tools “remain tools that work properly only in human hands, with a person always standing at the end of the analysis a model performs and verifying what the tool has done.” Marcin Dudek, head of CERT Polska, said the team’s experts need access to such tools to efficiently detect and repair vulnerabilities before attackers do.
What this means for AI safety in practice
As AI systems become more capable, international collaboration between developers and public research institutions will be essential to understanding their risks and building stronger safeguards. That is OpenAI’s stated reasoning, and it is the core idea behind the NASK agreement.
Nielek put the point directly: “AI safety cannot be achieved in isolation. It is a collaborative effort, and only through active cooperation between public AI Safety Institutes and frontier AI developers can we ensure that AI becomes truly reliable and safe to use.” The partnership, in his framing, should let the two sides anticipate risks earlier, strengthen safeguards and build public trust by combining NASK’s research expertise with OpenAI’s technical capabilities.
The practical meaning is threefold. First, evaluation: national institutes can test increasingly capable systems against the conditions of their own countries, languages and public services, catching problems that a single company’s internal testing might miss. Second, safeguards: the work of identifying vulnerabilities and strengthening protections becomes shared, drawing on research expertise from outside the company. Third, standards: common approaches to evaluating capabilities, assessing risks, testing safeguards and reporting incidents give countries a shared basis for comparing evidence — which matters because AI risks cross borders.
Reuters, reporting the announcement, noted the deal comes amid growing scrutiny of AI agents behaving in ways their developers did not intend. That scrutiny is the backdrop against which partnerships like this one will be judged: not by the signing of agreements, but by whether they produce the kind of verifiable results the CERT Polska pilot already has.
What is still open
For all its promise, the agreement leaves a great deal unanswered. The most important open question is what specific joint research projects will follow: no projects, timelines or deliverables have been announced. Readers should hold the new agreement to the July standard — named findings, fixed, confirmed — and judge it by what it produces.
It is also not yet clear whether NASK will gain broader access to OpenAI models beyond CERT Polska’s Daybreak access, or how evaluation findings from the collaboration will be shared or published. The agreement describes an intention to collaborate; the mechanisms for sharing results, and the question of who sees them, have not been stated.
Finally, the scope of the partnership is deliberately broad — AI safety, security and technical research — which means its success will depend on the specific work the two sides choose. The CERT Polska record shows what the model can deliver when it is focused on a concrete problem. The test for the OpenAI NASK partnership is whether the new agreement produces results of the same kind: specific vulnerabilities found, specific safeguards strengthened, specific evidence shared.
