OpenAI Signs AI Safety Research Deal With Poland's NASK, Whose CERT Polska Team Used OpenAI Models to Find Six MikroTik Flaws
OpenAI and Poland's NASK National Research Institute have signed an agreement to explore joint AI safety and security research through NASK's AI Safety Research Center, which serves as Poland's AI safety institute. Both sides point to earlier work: CERT Polska, which sits inside NASK, used OpenAI's cyber-capable models through Daybreak to help find six vulnerabilities in MikroTik RouterOS. The agreement's terms, funding and model access were not published.
This is a framework to "explore collaboration," not a funded program with published terms, but it gives OpenAI a formal research link to a European public AI safety institute and builds on a cyber-defense result both sides can point to. OpenAI and Poland's NASK National Research Institute announced the agreement on Friday, October 23.
What the agreement covers
OpenAI says the two organizations "have signed an agreement to explore collaboration on AI safety, security and technical research." On NASK's side, the work will be led by its AI Safety Research Center, which OpenAI says serves as the Polish Artificial Intelligence Safety Institute. OpenAI says the two will work on ways to evaluate increasingly capable AI systems, identify potential vulnerabilities and strengthen safeguards.
OpenAI places the deal alongside its existing work with the US Center for Advancing Standards and Innovation for Super Intelligence (CAISSI) and the UK AI Security Institute. It also repeats its call for the US to work with other countries on international technical standards for frontier AI.
NASK director Radosław Nielek said in OpenAI's announcement that "AI safety cannot be achieved in isolation." Tom Duff Gordon, OpenAI's vice president and head of EMEA policy, said the agreement "opens the door" to bringing Poland's research expertise "into the international effort to make AI safer."
In NASK's own Polish-language release, Nielek also calls it "the only such agreement in the European Union" (our translation). OpenAI's English announcement does not make that claim.
The CERT Polska precedent
Both announcements say the agreement builds on earlier work by CERT Polska, the national cyber incident team that operates within NASK. According to OpenAI, CERT Polska gained access to OpenAI's cyber-capable models through Daybreak in July. OpenAI says the team used the models to help discover six vulnerabilities in MikroTik RouterOS, including flaws that could be combined to take control of affected devices without authentication. OpenAI says the vendor released fixes, which CERT Polska confirmed prevented the attacks it observed.
Reuters framed the deal against recent scrutiny of OpenAI's safety practices, including the company's account of AI agents that escaped containment and attacked Hugging Face, which we covered in OpenAI's technical report on the incident.
Limits
- Neither announcement gives the agreement's length, funding, or what model access NASK will get.
- The MikroTik account comes from OpenAI and NASK. Neither names the six vulnerabilities or the affected RouterOS versions, and we did not match them to specific advisories.
- The "only such agreement in the European Union" line is NASK's claim, translated by us from Polish; we did not check it.
Sources
- OpenAI: OpenAI and Poland's NASK partner to advance AI safety and security (October 23, 2026)
- NASK: Jedyna taka umowa w UE. Partnerstwo NASK i OpenAI w obszarze bezpieczeństwa sztucznej inteligencji (in Polish, October 23, 2026)
- Reuters via Investing.com: OpenAI and Poland's NASK collaborate on AI safety research (October 23, 2026)
- Times of AI: OpenAI launches Daybreak (May 11, 2026)
- Times of AI: Hugging Face post-mortem (August 26, 2026)