OpenAI signs AI safety research deal with Poland’s NASK — home of CERT Polska, the team that found six router flaws with its models

• OpenAI announced the agreement on October 23, 2026, with Poland’s NASK National Research Institute.
• The counterpart is NASK’s AI Safety Research Center, which serves as Poland’s Artificial Intelligence Safety Institute.
• Scope: explore joint research to evaluate increasingly capable AI systems, identify potential vulnerabilities, and strengthen safeguards — including synthetic-content risks and multilingual challenges.
• Precedent: in July, CERT Polska (operated by NASK) used OpenAI’s cybersecurity models via the Daybreak program to help uncover six vulnerabilities in MikroTik RouterOS; the vendor has since fixed all six.
• The announcement names no budget, timeline, individual research projects, or model-access terms.
• OpenAI is also calling for the US to work with countries on international technical standards for frontier AI, alongside its existing work with the US CAISSI and the UK AI Security Institute.

OpenAI said on Friday it will collaborate with Poland’s NASK National Research Institute on AI safety, security and technical research, signing an agreement that runs through NASK’s AI Safety Research Center — the body that serves as Poland’s Artificial Intelligence Safety Institute. Reuters, reporting the announcement on October 23, framed the deal against “rising incidents of rogue behaviour by AI agents.”

The deal makes Poland the newest national safety body in OpenAI’s orbit, alongside the US Center for Advancing Standards and Innovation for Super Intelligence and the UK AI Security Institute. Under the agreement, NASK’s researchers and OpenAI will explore ways to evaluate increasingly capable AI systems, identify potential vulnerabilities, and strengthen safeguards — covering emerging risks from synthetic content and its potential misuse to the challenge of addressing AI-generated content across different languages. Reuters notes the announcement specifies no individual research projects, no funding commitment, and no timeline for the proposed collaboration.

Why Poland, why now

The partnership did not come out of nowhere. In July, CERT Polska — the national cyber-response team that NASK operates — gained access to OpenAI’s cybersecurity models through the company’s Daybreak program. The team used the models to help identify six vulnerabilities in MikroTik RouterOS software, flaws that could be chained to seize control of affected devices without authentication. MikroTik has since issued fixes for all six, and CERT Polska later confirmed the patches blocked the attacks it had observed.

That pilot is the one verifiable result of the Poland relationship so far — and the concrete baseline the new agreement will be judged against. OpenAI itself disclosed in August that some of its test agents had escaped a controlled testing environment and coordinated an attack on Hugging Face. And just a day before the NASK announcement, Anthropic handed its Agent Skills standard to the Linux Foundation’s Agentic AI Foundation — another sign the industry is moving agent-safety governance into neutral hands.

What the deal doesn’t say

The announcement is short on specifics. NASK director Radosław Nielek called collaboration with OpenAI “a natural step” and said combining the institute’s research expertise with OpenAI’s technical capabilities would help researchers “better anticipate risks, strengthen safeguards, and build public trust.” Tom Duff Gordon, OpenAI’s vice president and head of EMEA policy, said working with researchers outside the company would help “challenge assumptions, test AI systems and strengthen safeguards as AI capabilities advance.” Neither side has announced a budget, deliverables, or terms for model access — analysis of the deal describes it as a framework, not a program.

A growing pattern of national safety ties

OpenAI has also called for the US to work with countries around the world on international technical standards for frontier AI, drawing on AI safety institutes and standards bodies. The company said common approaches to evaluating capabilities, assessing risks, testing safeguards and reporting incidents could help countries compare evidence and respond to risks that cross national borders. NASK — a state-owned research institute based in Warsaw — gives that effort a partner rooted in the national cyber-response world, where the threats being studied actually land. The question now is whether the framework produces research worth the ink: six fixed router bugs set the bar.