OpenAI and Poland's NASK sign an AI safety partnership: a new EU-side institute partner

On 2026-10-23, OpenAI and Poland's research institute NASK announced an AI safety research partnership. OpenAI's official news RSS carried the item at 00:00 GMT under the title "OpenAI and Poland's NASK partner to advance AI safety and security," with the description "OpenAI and Poland's NASK are partnering on AI safety, security, and technical research to strengthen safeguards and support a more resilient digital ecosystem" (the article page returned 403 to this site, so the RSS description and NASK's own full announcement are the primary sources here). NASK's same-day announcement is blunter: its headline reads Jedyna taka umowa w UE — the only such agreement in the EU.

What the agreement is

In NASK's wording: new threats require joint action, and what the two organizations signed is a cooperation agreement on AI safety research. It combines their knowledge and experience to identify threats related to this technology faster, protect against them better, and more effectively verify how advanced AI methods can help increase the security and resilience of digital services.

The announcement gives no project list, funding figures, or timeline — what it does give is the framework and the named leads on both sides.

Who leads each side

On the NASK side, the lead is the Ośrodek Badań nad Bezpieczeństwem Sztucznej Inteligencji (AI Security Research Center). The announcement states plainly that this center performs the role of Poland's AI Security Institute (AISI) and will research the safety of increasingly advanced AI systems together with OpenAI; the analyses it conducts are meant to help evaluate those systems, detect their weak points, and strengthen safeguards.

The OpenAI-side quote comes from Tom Duff Gordon, VP and head of public policy for EMEA. His point, in substance: Poland has excellent scientists and distinctive technology talent, and they already take part in research at OpenAI; as AI capabilities grow, an outside perspective is needed to verify assumptions, test systems, and strengthen safeguards — and the agreement with NASK makes that a joint effort, bringing Polish research contributions into international AI safety work.

NASK director Radoslaw Nielek's quote describes the center's day-to-day research surface: studying new threats, developing safeguards, detecting AI-generated content, and examining how AI can be turned to harmful use — and how those threats differ depending on the language of the conversation. He calls the agreement an important step and frames the EU-only distinction as recognition of his team's specialists.

Where the EU-only claim comes from

Jedyna taka umowa w UE is NASK's own headline, and the same phrase appears in Nielek's quote — "the only such agreement in the European Union, which is an important distinction and recognition of our specialists' competence for us." OpenAI's RSS description makes no such comparison.

This site reports it as an attribution: it is one signatory's characterization, not an independently verified census. What other arrangements exist between EU-based AI safety institutes and model developers is not something this article audits.

Where it sits in OpenAI's institute sequence

The announcement places the signing in an existing series: OpenAI's collaborations with the US Center for Advancing Standards and Innovation for Super Intelligence (CAISSI), the UK AI Security Institute (AISI), and other public research institutions "helped detect and remove vulnerabilities in AI agents and models" and contributed to improving the safety of the most advanced models — followed by the line that Poland now joins that group.

A separate section covers standards (Bezpieczna AI wymaga wspólnych zasad — safe AI requires common rules): OpenAI has advocated that the United States cooperate with other countries on international technical standards for the most advanced AI systems, with AI safety institutes and standardization organizations helping to develop them. Common rules for evaluating AI capabilities, analyzing risk, testing safeguards, and reporting incidents would let countries compare results and respond to threats across borders. NASK's contribution is the Polish AI safety research knowledge it uses daily to protect public services and the Polish information space; the announcement argues that national partners help standards respond to concrete threats and local conditions of AI use.

The backstory: Daybreak and the six MikroTik vulnerabilities

The signing did not start from zero — the announcement itself supplies a checkable history: in July, CERT Polska, the incident response team operating within NASK, received access to advanced OpenAI models for cybersecurity work through the Daybreak program. Using that access, the team found six vulnerabilities in MikroTik RouterOS; chaining some of them allowed taking control of a device without logging in. The vendor shipped patches, and CERT Polska confirmed the patches block the demonstrated attack methods.

That backstory is worth attention because it shows what these collaborations actually produce: not a letter of intent, but a closed loop of vulnerability discovery and vendor patches.

What it means for ChatGPT and API users

This announcement lives at the policy and safety-research level: no product changes, no model availability changes, no pricing changes. For readers, the substance is in three places — OpenAI's institute network gains an EU-side member (the only one in the EU, per NASK's own framing); Poland's counterpart is the center performing the national AISI role; and Daybreak-style access programs have another documented instance of concrete output. If safety-evaluation or standards work produces tangible results, this site will follow up.

Next steps

• For OpenAI's current model family and version timeline, see The GPT-6 family grows: gpt-6-sol, gpt-6-luna, gpt-6.1-sol land in the official SDKs;
• For another recent policy-and-safety data point, see OpenAI's misalignment report: an internal research model used a funded test crypto wallet without authorization;
• More updates in the news channel.