OpenAI has partnered with Poland’s NASK National Research Institute to strengthen artificial intelligence safety research, bringing together expertise in advanced AI systems, cybersecurity, and emerging digital threats.
The OpenAI NASK AI safety partnership aims to improve how researchers evaluate increasingly capable AI models, identify potential security weaknesses, and develop safeguards against malicious applications of the technology.
The agreement reflects growing concerns about the security implications of advanced AI, particularly as autonomous systems become more capable of performing complex tasks with limited human supervision.
For Poland, the collaboration represents an opportunity to play a greater role in international AI safety research. OpenAI, meanwhile, gains access to additional cybersecurity expertise and independent research perspectives.
OpenAI and NASK Expand Research Into Advanced AI Security
The partnership brings OpenAI together with NASK’s AI Safety Research Center, which supports Poland’s efforts to investigate the risks associated with increasingly sophisticated artificial intelligence systems.
Their research will examine how advanced AI models behave under different conditions, where vulnerabilities may emerge, and how existing safeguards can protect against misuse. Particular attention will go toward cybersecurity threats, AI-generated misinformation, and potential weaknesses in autonomous systems.
NASK already has experience investigating digital security incidents and identifying threats to public infrastructure. Combining that expertise with OpenAI’s model development capabilities could help researchers better understand how powerful AI systems interact with real-world security environments.
The agreement also creates opportunities for closer technical exchanges between researchers. Rather than treating AI safety as a responsibility belonging exclusively to technology companies, the collaboration brings a national research institution directly into the process.
Previous Cybersecurity Research Uncovered Six RouterOS Vulnerabilities
OpenAI and NASK are not starting their relationship from scratch. Their earlier cybersecurity collaboration demonstrated how advanced AI models could support researchers investigating vulnerabilities in widely used software.
In July 2026, CERT Polska, which operates within NASK, gained access to OpenAI’s cybersecurity-capable models through the company’s Daybreak initiative. Researchers subsequently used the technology to examine MikroTik RouterOS, an operating system used in networking equipment.
The investigation identified six security vulnerabilities, including weaknesses that could potentially allow attackers to compromise affected devices without authentication when combined under certain conditions.
MikroTik later released patches addressing the reported issues, while CERT Polska verified that the fixes prevented the identified attack methods.
The findings offered a practical demonstration of AI-assisted cybersecurity research. Instead of replacing security specialists, advanced models helped researchers investigate complex software vulnerabilities and accelerate parts of the discovery process.
That earlier work provides a foundation for the new partnership, particularly as both organisations explore how AI can strengthen defensive security capabilities.
Poland Takes a Larger Role in Global AI Safety Research
Poland’s involvement reflects its growing interest in contributing to international efforts to develop safer artificial intelligence systems.
NASK has extensive experience protecting digital infrastructure, responding to cybersecurity incidents, and researching threats involving online manipulation and misinformation. Its work places the institute in a position to examine AI risks that extend beyond traditional software vulnerabilities.
The organisation also investigates synthetic media and AI-generated content, areas that have become increasingly important as generative AI tools make convincing digital manipulation easier and more accessible.
Through the agreement, Polish researchers will have additional opportunities to contribute their expertise to the evaluation of advanced AI systems.
The collaboration could also strengthen connections between Poland’s cybersecurity community and international AI developers, helping local researchers participate in discussions about technical safeguards, security assessments, and responsible deployment.
OpenAI Supports International Cooperation on AI Safety Standards
OpenAI’s agreement with NASK forms part of a broader effort to strengthen international cooperation around the evaluation and security of advanced artificial intelligence.
The company has pursued similar research relationships with organisations working on AI security and technical assessments in other countries, including the United Kingdom and the United States.
Such partnerships are becoming more important as AI systems grow more powerful and their potential risks become increasingly difficult for individual organisations to assess independently.
Different research institutions bring their own testing methods, technical experience, and understanding of regional threats. Working together can help identify weaknesses that might otherwise escape detection during internal evaluations.
International cooperation could also support the development of more consistent approaches to AI safety testing, particularly in areas involving cybersecurity, model behaviour, and the detection of malicious applications.
For OpenAI, expanding these relationships provides additional opportunities to evaluate its technology against different security challenges while contributing to discussions about shared safety practices.
Autonomous AI Systems Create New Cybersecurity Challenges
The rapid development of autonomous AI agents has introduced another layer of complexity to cybersecurity research.
Unlike conventional chatbots that primarily generate responses to user prompts, advanced AI agents can interact with software tools, execute instructions, analyse information, and perform sequences of tasks with varying degrees of independence.
These capabilities create opportunities for productivity and automation, but they also introduce risks when systems encounter malicious instructions, unreliable information, or unexpected operating conditions.
Researchers must consider whether AI agents can recognise harmful requests, maintain appropriate security boundaries, and avoid taking unintended actions when interacting with external environments.
Traditional cybersecurity testing remains important, although evaluating autonomous AI requires additional attention to model behaviour and decision-making.
The OpenAI–NASK partnership provides another avenue for investigating these challenges through technical research and practical security assessments.
What the OpenAI–NASK Partnership Means for the Future of AI Security
The agreement highlights an increasingly important connection between artificial intelligence development and cybersecurity research.
OpenAI’s earlier work with CERT Polska demonstrated how advanced AI models could assist security specialists in discovering vulnerabilities in existing software. The new partnership creates opportunities to build on that experience while examining broader questions surrounding AI safety.
For NASK, the collaboration could strengthen its contribution to international AI security research and provide greater exposure to emerging technologies that may influence future cybersecurity threats.
OpenAI also stands to benefit from research conducted outside its own development teams, particularly when evaluating how advanced systems respond to unfamiliar threats and real-world security conditions.
The partnership remains an early step, and its long-term impact will depend on the research findings and practical improvements that emerge from the collaboration.
As AI capabilities continue to expand, cooperation between technology developers and independent research institutions could become increasingly important in identifying security weaknesses before they develop into larger problems.
