OpenAI Partners with NASK for AI Security in Poland

OpenAI and the Polish public institute NASK have entered into a collaboration agreement focused on the safety and security of AI. This step follows work conducted with CERT Polska, which led to patches for MikroTik RouterOS, and is part of an international effort towards common technical standards.

Tests conducted by CERT Polska led to patches

Before the agreement was signed, OpenAI was already collaborating with cybersecurity teams in Poland. In July, CERT Polska, affiliated with NASK, was able to access OpenAI's cybersecurity models through Daybreak. The team used these models to help identify six vulnerabilities in MikroTik RouterOS, some of which could be combined to allow for the takeover of devices without authentication. The provider of MikroTik RouterOS released corrective updates, validated by CERT Polska as having blocked the identified attacks.

What the OpenAI–NASK agreement on AI security entails

OpenAI and NASK have signed an agreement aimed at exploring cooperation around security, safety, and technical research in artificial intelligence. NASK's AI Security Research Center, which serves as the Polish Institute for AI Security, and OpenAI will work together to analyze increasingly advanced AI systems, detect potential vulnerabilities, and strengthen protective measures. The agreement also includes analyzing how advanced AI systems can contribute to making the digital ecosystem safer and more resilient.

Priorities outlined by NASK and OpenAI

For Radosław Nielek, director of NASK, this cooperation continues ongoing research, including the development of protective measures against emerging risks, detection of AI-generated content, analysis of malicious uses, and consideration of multilingual aspects. He believes that AI security requires active cooperation between public institutes and leading AI developers to ensure the reliability and safety of systems. Tom Duff Gordon, Vice President and Head of EMEA Policy at OpenAI, highlights the presence of scientific and technical talent in Poland, some of whom have contributed to OpenAI's research. He considers it essential to rely on external expertise to challenge assumptions, test systems, and strengthen protections. According to him, the agreement with NASK allows for the integration of Polish expertise into an international effort to make AI safer.

Towards common approaches and technical standards

OpenAI calls on the United States to collaborate with other countries to establish international technical standards for advanced AI, leveraging the expertise of AI security institutes and standardization bodies. Common methods for assessing capabilities, analyzing risks, testing protective measures, and reporting incidents are presented as ways to help countries compare evidence and respond to risks that cross borders. The integration of national expertise, such as that of NASK, is emphasized. NASK bases its know-how on in-depth work in AI security, aimed at protecting public services and the Polish digital landscape. Without national actors like NASK, international standards for AI security may not take into account practical uses. OpenAI specifies that its collaborations with public bodies have contributed to improving protective measures, identifying and correcting vulnerabilities in AI agents, and advancing AI security at the highest level.