OpenAI Partners With Poland’s NASK to Bolster AI Safety Research

The partnership leverages the expertise of the NASK AI Safety Research Center, which currently functions as Poland’s national AI safety institute. A primary objective is to replicate the success of recent security testing conducted by CERT Polska. In July, the incident response team utilized OpenAI’s 'Daybreak' model access to identify six critical vulnerabilities within MikroTik RouterOS. These flaws, which allowed for unauthorized system control, were subsequently patched by the vendor after the team validated the attack paths.

Beyond cybersecurity, the collaboration aims to address the complexities of synthetic content and multilingual AI risks. Radosław Nielek, Director of NASK, emphasized that AI safety requires active cooperation between public institutes and private developers to ensure reliability. This initiative mirrors OpenAI’s existing frameworks with the US and UK safety institutes, focusing on four core pillars: capability evaluation, risk assessment, safeguard testing, and standardized incident reporting to track cross-border threats.