The Pulse
OpenAI and Poland’s NASK Sign AI Safety Research Pact
OpenAI and Poland’s NASK signed an agreement on October 23, 2026, to explore joint AI safety, security and technical research. The work will examine how to evaluate advanced AI systems, find vulnerabilities and strengthen safeguards, buildi
AI.info Team · Written with AI · fact-checked ·
OpenAI wants countries to compare AI safety tests using common methods; Poland’s NASK puts local risks, including synthetic content and multilingual misuse, in the foreground. Their new agreement brings those priorities into one collaboration, but its public terms describe a research agenda rather than a schedule of concrete tests or deliverables.
Announced October 23, 2026, the agreement commits OpenAI and NASK’s AI Safety Research Center to explore work on AI safety, security and technical research. The organizations say they will examine ways to evaluate increasingly capable systems, identify vulnerabilities and strengthen safeguards. They have not announced a timetable, named evaluations or described how results will be shared.
Common tests meet local risks
OpenAI frames the partnership as part of a broader effort to establish shared technical approaches to measuring capabilities, assessing risks, testing safeguards and reporting incidents. Common methods, the company argues, could help governments compare evidence and respond to risks that cross borders. OpenAI points to work with the U.S. Center for Advancing Standards and Innovation for Super Intelligence and the UK AI Security Institute as part of that wider effort.
NASK’s research center serves as Poland’s Artificial Intelligence Safety Institute. Its director, Radosław Nielek, names a wider set of concerns: safeguards against emerging risks, detection of AI-generated content, malicious uses and the challenges posed by multiple languages. “AI safety cannot be achieved in isolation,” Nielek says. He argues that cooperation between public safety institutes and AI developers is needed to test systems and build safeguards.
NASK’s focus extends beyond model evaluations
The two organizations’ emphases are complementary, but they raise a practical question: can shared evaluations account for different languages and the public services or information systems where AI is used? OpenAI says national partners can help connect broad standards to those conditions. NASK’s stated focus on multilingual content and malicious uses gives the collaboration a local research agenda alongside its work on system evaluation.
Tom Duff Gordon, OpenAI’s vice president and head of EMEA policy, says the agreement will bring Polish research expertise into international safety work. “This agreement with NASK opens the door to doing that together, bringing Poland’s research expertise into the international effort to make AI safer,” he says. The announcement does not identify specific research projects or say whether the partners will publish shared findings.
CERT Polska’s RouterOS work provides a test case
The agreement follows practical cybersecurity work between OpenAI and defenders in Poland. OpenAI says CERT Polska, which operates within NASK, received access to its cyber-capable models through the Daybreak initiative in July. The team used the models to help identify six vulnerabilities in MikroTik RouterOS; some flaws could be combined to take control of affected devices without authentication.
MikroTik released fixes, and CERT Polska confirmed they prevented the attacks it observed, according to OpenAI. NASK’s July account separately describes its access to OpenAI’s GPT-5.5 Cyber model and the role of human specialists in reviewing model outputs. That earlier work offers a concrete example of the security research behind the new agreement; the October announcement broadens the remit to AI safety research, but leaves its next projects and milestones unspecified.