OpenAI and Poland's NASK National Research Institute signed an agreement on October 23, 2026 to explore collaboration on AI safety, security and technical research. The work pairs NASK's AI Safety Research Center, which acts as the Polish Artificial Intelligence Safety Institute, with OpenAI teams evaluating capable systems and safeguards. The deal builds on July 2026 cyber cooperation in which CERT Polska used OpenAI models to help find six vulnerabilities in MikroTik RouterOS. For business, the signal is that national safety institutes are becoming direct partners in setting testing practice.
How the NASK and OpenAI research pact is structured
Under the agreement, the two organizations will jointly study how to evaluate increasingly capable AI systems, identify potential vulnerabilities and strengthen safeguards. OpenAI said the goal is to better understand emerging risks and explore how advanced AI can support a more secure and resilient digital ecosystem. NASK Director Radoslaw Nielek said the center works on safeguards against emerging risks, detection of synthetic AI-generated content, malicious uses of AI and multilingual aspects of these challenges. Tom Duff Gordon, vice president and head of EMEA policy at OpenAI, said outside expertise is needed to challenge assumptions and test systems as capabilities grow.
OpenAI framed the partnership as an extension of its work with the US Center for Advancing Standards and Innovation for Super Intelligence, the UK AI Security Institute and other public research bodies. That earlier cooperation, according to the company, helped strengthen safeguards, identify and fix security vulnerabilities in AI agents and advance frontier safety research. Nielek said safety cannot be achieved in isolation and only cooperation between public institutes and frontier developers can make AI reliable and safe to use. Gordon added that Poland has exceptional scientists and technical talent, many of whom have helped lead OpenAI research, bringing national expertise into the international safety effort.
The announcement connects the NASK deal to OpenAI's September 21, 2026 call for US-led global technical standards for frontier AI, including recursive self-improvement where AI systems drive development of successive generations. OpenAI named three problems for standards to address: fragmentation of evaluations and incident definitions, collective action failures when nations act alone, and uneven capacity because frontier work is not evenly distributed. It cited safety institutes in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India and the United Kingdom as a network for standard-setting. Without national partners like NASK, the company said, global standards may stay detached from real conditions of AI use.
What the safety and cyber cooperation means for business
For companies deploying AI, closer ties between developers and national institutes point to more comparable methods for capability evaluation, risk assessment, safeguard testing and incident reporting. OpenAI proposed standards for measuring AI progress relevant to recursive self-improvement, for human oversight of automated research with triggers for immediate review, and for incident classification with common severity levels and reporting thresholds. The company said such standards would not be licenses or mandatory prerelease review of models, with governments deciding on legal adoption. Large firms operating across borders would feel harmonized reporting first, while smaller firms would later inherit clearer testing expectations from vendors.
The Polish cyber track record shows what such access can produce in practice, along with its limits. In July 2026, CERT Polska, which operates within NASK, gained access to cyber-capable models through Daybreak, and used them to help discover six flaws in MikroTik RouterOS that could be combined for unauthenticated control; the vendor released fixes that CERT Polska confirmed blocked the observed attacks. NASK's July 2 announcement said it received GPT-5.5 Cyber through the Government Trusted Access for Cyber program, calling Poland the third European country with such access, for malware analysis, forensics and vulnerability discovery. Nielek stressed that a person always verifies model output, while CERT Polska head Marcin Dudek said experts need such tools to find flaws before attackers do.
The marker to watch is whether joint NASK-OpenAI work feeds into shared evaluation methods, safeguard tests or incident-reporting thresholds referenced by other safety institutes. Further steps would include published test results, common severity scales, or government decisions to incorporate voluntary technical norms into national rules. If national institutes begin citing the same evidence base, vendors will face more consistent demands on documentation and monitoring. Absence of such outputs would leave the agreement as bilateral research without wider operational effect.
