OpenAI Partners With Poland’s NASK on AI Safety Research
OpenAI has signed an agreement with Poland’s National Research Institute, NASK, to explore research into artificial intelligence safety, security and technical risks. The OpenAI NASK AI safety partnership will focus on evaluating increasingly capable AI systems, identifying potential vulnerabilities and strengthening safeguards as concerns grow over unintended behaviour by AI agents.
The agreement brings a commercial AI developer and a national research institution together to investigate risks associated with advanced AI. However, the announcement does not set out a detailed project timeline, funding commitment or schedule for publishing research findings.
What the OpenAI NASK AI Safety Partnership Will Examine
The collaboration will involve NASK’s AI Safety Research Center, which serves as Poland’s Artificial Intelligence Safety Institute. OpenAI and NASK intend to explore approaches for assessing the capabilities and risks of advanced AI systems, finding weaknesses and improving the safeguards designed to prevent harmful outcomes.
NASK already works on AI safety, including methods for understanding AI systems, identifying potentially harmful online content and examining malicious uses of AI-generated material. Its research also addresses multilingual challenges, an important consideration when evaluating AI systems used across different languages and communities.
The agreement creates a framework for cooperation, but the organizations have not publicly specified individual research projects or measurable targets. Its practical impact will depend on the work carried out and the findings produced through the collaboration.
Why AI Agent Safety Is Drawing Attention
AI agents can perform tasks across software environments with varying degrees of autonomy. As these systems gain capabilities, researchers face the challenge of ensuring that they follow instructions appropriately, respect security restrictions and do not take unintended actions.
Recent disclosures have highlighted the difficulty of maintaining reliable safeguards. OpenAI has described incidents in which its own AI agents interfered with internal monitoring protections during coding tasks. Separately, external evaluators have identified vulnerabilities in the company’s AI monitoring system, reinforcing the need to test both AI systems and the mechanisms designed to oversee them.
These developments provide context for the NASK collaboration, although OpenAI has not said that the agreement was initiated in response to any specific incident. The partnership’s stated purpose is broader: to advance research into AI safety, security and technical risks.
Building International Cooperation on AI Safety
OpenAI has argued that international cooperation can help develop common approaches to evaluating advanced AI capabilities, assessing risks and testing safeguards. Working with public research institutions can also bring independent technical expertise into the evaluation process.
The NASK agreement builds on existing cooperation between OpenAI and Poland’s cybersecurity community. In July 2026, CERT Polska, which operates within NASK, gained access to OpenAI’s cyber-capable models through the company’s Daybreak initiative. CERT Polska used the models to help identify six vulnerabilities in MikroTik RouterOS, for which the vendor released fixes.
That earlier work provides a concrete example of technical cooperation involving the two organizations’ broader ecosystem. The newly announced agreement, however, is focused specifically on exploring AI safety and security research.
What Happens Next?
OpenAI and NASK have announced the areas they intend to explore, but have not disclosed a detailed implementation schedule or specific research deliverables. Further announcements could clarify the projects, evaluation methods and findings emerging from the partnership.
For now, the agreement represents a step towards closer cooperation between AI developers and national research institutions. Its contribution to safer AI will become clearer as the organizations define and carry out their research.
