OpenAI and Poland’s NASK Partner to Advance AI Safety Research
An agreement to investigate cooperation on AI safety, security, and technical research has been signed by OpenAI and Poland’s NASK National Research Institute. To better identify new threats, bolster defenses, and investigate how cutting-edge AI can support a more secure and resilient digital environment, the partnership will bring together expertise from both organizations.
International cooperation between developers and public research institutes will be critical as AI systems grow more powerful to understand their hazards and create more robust defenses. The US Center for Advancing Standards and Innovation for Super Intelligence (CAISSI), the UK AI Security Institute (AISI), and other public research organizations worldwide have already benefited from OpenAI’s ongoing efforts to improve security, find and address security flaws in AI agents, and advance frontier AI safety.
Through the agreement, OpenAI and NASK’s AI Safety Research Center, which functions as the Polish Artificial Intelligence Safety Institute (AISI), will work together on AI safety research, investigating methods to assess increasingly powerful AI systems, spot potential weaknesses, and bolster defenses.
Using the knowledge of AI safety institutes and standards agencies, OpenAI has urged the US to collaborate with nations worldwide on international technical standards for frontier AI. Countries can exchange evidence and respond to cross-border threats by using common methods for assessing capabilities, evaluating risks, testing protections, and reporting events. The partnership with NASK demonstrates the need to incorporate national knowledge into this greater effort.
In NASK’s case, this knowledge is based on in-depth AI safety research that is directly applied to safeguarding Poland’s digital information ecosystem and public services. Global AI safety standards might continue to be disconnected from the real-world applications of AI in the absence of national partners like NASK.
Also, the arrangement expands on OpenAI’s current collaboration with Polish cyber defenders. Through Daybreak, CERT Polska, a division of NASK, was able to access OpenAI’s cyber-capable models in July. Six vulnerabilities in MikroTik RouterOS, including ones that might be combined to take control of impacted devices without authentication, were found by the researchers using OpenAI models. CERT Polska verified that the vendor’s updates averted the attacks it observed.
