OpenAI and Poland's NASK partner on AI safety research

A new research alliance

OpenAI has entered into a formal agreement with Poland’s NASK to conduct joint research on AI safety and security. This collaboration provides Polish public researchers with direct access to frontier models and the specific testing workflows used by the company. The goal is to create a structured environment where researchers can evaluate the risks of increasingly capable systems.

NASK’s AI Safety Research Center serves as the Polish Artificial Intelligence Safety Institute (AISI). By partnering with a major developer like OpenAI, the institute can better identify potential vulnerabilities in new technologies. This move is intended to strengthen the safeguards that prevent AI from being used in harmful ways or causing unintended consequences.

The agreement highlights a growing trend of collaboration between private AI developers and national safety bodies. OpenAI noted that many of the scientists who have led its own internal safety work are originally from Poland. This cultural and professional connection provides a foundation for the new technical partnership between the two entities.

Testing in the real world

The partnership is already rooted in practical experience involving CERT Polska, which is the incident response team within NASK. In July, this team gained access to OpenAI’s cyber-capable models through a specialized channel called Daybreak. Analysts used these models to probe MikroTik RouterOS and discovered six distinct vulnerabilities.

These flaws were significant because they could be chained together to allow an attacker to take control of affected devices without any authentication. Following the discovery, MikroTik released official fixes for the RouterOS flaws. CERT Polska then confirmed that these patches successfully prevented the specific attacks their analysts had recorded against unpatched hardware.

This cycle of discovery, remediation, and validation is the primary model OpenAI wants to repeat with national institutes. By allowing researchers to point models at specific infrastructure, the companies can find and fix bugs before they are exploited by malicious actors. This iterative loop ensures that security updates are grounded in actual testing rather than just theoretical risks.

Focus on public services

NASK plans to apply its safety research specifically to the protection of public services and the broader digital information environment in Poland. A major part of this remit involves detecting synthetic AI-generated content. The center will work to track how such material is used maliciously across various different languages.

The multilingual aspect of this research is particularly important for Poland, where public messaging is not restricted to a single language. By understanding how synthetic content spreads in different tongues, researchers can better defend against disinformation campaigns. This work helps ensure that AI-generated media does not undermine the integrity of public communication.

Radosław Nielek, Director of NASK, stated that the collaboration is a natural step for the institute to develop safeguards against emerging risks. He emphasized that AI safety cannot be achieved in isolation and requires active cooperation between public institutes and frontier developers. This joint effort aims to make AI systems more reliable and safe for general use across the public sector.

Building global standards

OpenAI is positioning this partnership as part of a broader international effort to establish technical standards for frontier AI. The company already works with the UK AI Security Institute (AISI) and the US Center for Advancing Standards and Innovation for Super Intelligence (CAISSI). These existing relationships have already helped identify and fix security vulnerabilities in AI agents.

OpenAI has called for increased coordination between the United States and other countries to create international technical standards. They want to draw on the expertise of various safety institutes and standards bodies to ensure consistency. By involving Poland, OpenAI is expanding the pool of international experts who can contribute to these global rules.

The company has outlined four shared activities for these national institutes to follow. These include capability evaluation and risk assessment as the primary steps. These are followed by safeguard testing and incident reporting, which allow different countries to compare evidence when a specific risk crosses a national border.

The path forward

The collaboration between OpenAI and NASK represents a move toward a more transparent and collaborative safety model. Instead of safety being a closed process, it becomes a shared responsibility between the people who build the models and the public researchers who test them. This approach allows for faster identification of bugs and more robust defenses against cyber threats.

For Polish researchers, this means a seat at the table in shaping how the world interacts with frontier AI. They will be able to contribute to the technical details of how models are secured and how risks are mitigated. This expertise will likely influence how AI is deployed in public services and infrastructure throughout the region.

As AI capabilities continue to increase, the need for these types of partnerships will likely grow. The focus will remain on creating a reliable environment where AI can be used without posing significant risks to security or public information. The success of the NASK partnership will serve as a blueprint for how other countries might engage with frontier AI developers.