Polish Research Institute NASK, OpenAI Reach Deal on AI Safety Work

OpenAI and Poland’s NASK National Research Institute have signed an agreement to explore collaboration on AI safety, security and technical research, OpenAI announced on October 23, 2026.

OpenAI said the partnership will bring together expertise from both organizations to better understand emerging risks, strengthen safeguards, and explore how advanced AI can contribute to a more secure and resilient digital ecosystem. Under the agreement, NASK’s AI Safety Research Center, which serves as the Polish Artificial Intelligence Safety Institute, and OpenAI will collaborate on AI safety research, exploring ways to evaluate increasingly capable AI systems, identify potential vulnerabilities and strengthen safeguards.

OpenAI said its existing work with the US Center for Advancing Standards and Innovation for Super Intelligence, the UK AI Security Institute and other public research institutions has already helped strengthen safeguards, identify and fix security vulnerabilities in AI agents, and advance frontier AI safety.

Statements From NASK and OpenAI Leadership

NASK Director Radosław Nielek said collaboration with OpenAI is a natural step for the institute and its AI Safety Research Center. He said the center’s work spans developing safeguards against emerging risks, detecting synthetic AI-generated content, and understanding both malicious uses of AI and the multilingual aspects of these challenges. AI safety cannot be achieved in isolation, he said, and only cooperation between public AI safety institutes and frontier AI developers can make AI truly reliable and safe to use. By combining NASK’s research expertise with OpenAI’s technical capabilities, he said, the two organizations can better anticipate risks, strengthen safeguards and build public trust in the technologies.

Tom Duff Gordon, vice president and head of EMEA policy at OpenAI, said Poland is home to exceptional scientists and technical talent, many of whom have helped lead OpenAI’s own research. He said that as AI becomes more capable it is vital to draw on expertise beyond OpenAI’s own organization to challenge its assumptions, test its systems and strengthen safeguards, and that the agreement with NASK opens the door to doing that together while bringing Poland’s research expertise into the international effort to make AI safer.

A Wider Push for International AI Standards

OpenAI said the NASK collaboration reflects the value of bringing national expertise into its wider call for international technical standards for frontier AI. The company has urged the United States to work with countries worldwide on such standards, drawing on AI safety institutes and standards bodies, and said common approaches to evaluating capabilities, assessing risks, testing safeguards and reporting incidents can help countries compare evidence and respond to risks that cross borders. Without national partners like NASK, OpenAI said, global AI safety standards may remain detached from the realities in which AI is actually used.

In a September 21, 2026 post, OpenAI said the United States should lead an effort to develop global technical standards for frontier AI, including for recursive self-improvement, a process in which AI systems increasingly drive development of successive generations of AI. The post named three challenges such standards would address: fragmentation, where different nations’ evaluations, reporting requirements and incident definitions could conflict; collective action, where nations acting independently can produce outcomes no nation wants; and uneven capacity, because frontier development activity and related expertise are not evenly distributed across the world.

The post pointed to AI safety institutes already established in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India and the United Kingdom as a network that could facilitate standard-setting.

It proposed standards for evaluating AI progress relevant to recursive self-improvement and the amount of autonomous research inside an AI company; for human oversight of automated AI research, including which automated research processes should trigger immediate human review; and for incident classification, tracking, reporting and response, such as common severity levels and reporting thresholds. OpenAI said these technical standards would not be licenses, mandatory prerelease review or approval requirements for AI models, and that national governments would decide whether and how to incorporate them into their own legal systems.

Earlier Cybersecurity Work in Poland

The agreement also builds on OpenAI’s existing work with cyber defenders in Poland. In July 2026, CERT Polska, which operates within NASK, gained access to OpenAI’s cyber-capable models through Daybreak. OpenAI said the CERT Polska team used its models to help discover six vulnerabilities in MikroTik RouterOS, including flaws that could be combined to take control of affected devices without authentication. The vendor released fixes, which CERT Polska confirmed prevented the attacks it observed, according to OpenAI’s account.

NASK’s own July 2, 2026 announcement said the institute gained access to GPT-5.5 Cyber, a specialized AI model created by OpenAI for detecting and analyzing cyber threats, through the Government Trusted Access for Cyber program for government units. NASK characterized Poland as the third European country, and one of few worldwide, to receive such access. It described Daybreak as an initiative to keep defenders a step ahead of criminals in using AI for cybersecurity, with GPT-5.5 Cyber applied to malware analysis, digital forensics and vulnerability discovery to accelerate the full cycle of detecting and repairing vulnerabilities.

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. CERT Polska head Marcin Dudek said the team’s experts need access to such tools to efficiently detect and repair vulnerabilities before attackers do, and added that CERT Polska already shares its knowledge from using large language models on its website and plans to continue.