By Hamza Faisal
OpenAI is taking its AI safety diplomacy to Warsaw. On Thursday, the lab announced it has signed an agreement with NASK, Poland's national research institute, to explore joint work on AI safety, security and technical research — the latest in a string of pacts linking frontier labs with national safety bodies.
The counterpart inside NASK is its AI Safety Research Center, which serves as Poland's AI Safety Institute. OpenAI frames the deal as "an agreement to explore collaboration," with work aimed at evaluating increasingly capable AI systems, finding vulnerabilities and strengthening safeguards. No budget, timeline or named deliverables have been made public.
A concrete precedent: six router flaws found with AI help
The announcement's most persuasive evidence is a July pilot. CERT Polska, the national incident-response team operating within NASK, was given access to OpenAI's cyber-capable models through Daybreak, the program through which OpenAI supplies vetted defenders with stronger security tooling.
OpenAI says the CERT Polska team used those models to help discover six vulnerabilities in MikroTik RouterOS, the firmware running millions of routers worldwide. Independent reporting describes the exploit chain, dubbed MikroTrick, as including a CVSS 9.2 SSH authentication bypass caused by incomplete RSA public-key validation, a second CVSS 9.2 flaw that turns a crafted username into full administrator privileges, and a CVSS 8.8 bandwidth-test flaw that can leak kernel memory or crash a device. Chained together, they could take control of affected devices without authentication.
The vendor has shipped fixes, and CERT Polska confirmed the fixes prevented the attacks it observed. Exploitation attempts were reportedly spotted from at least September 2 against devices exposing SSH to the public internet — a reminder for anyone running MikroTik gear to confirm a fixed RouterOS release and keep SSH off the open internet.
Why Poland, and why now
NASK director Radosław Nielek described the work as a natural step for an institute already active in AI safety research, listing three strands: safeguards against emerging risks, detecting synthetic AI-generated content, and understanding the multilingual aspects of misuse. The last point is the most Poland-specific: safety evaluations built mostly in English can miss failures that appear only in other languages, a gap that matters for protecting a country's public services and information space.
Tom Duff Gordon, OpenAI's VP and head of EMEA policy, called Poland "home to exceptional scientists and technical talent, many of whom have helped lead OpenAI's own research," adding that outside expertise is needed "to challenge our assumptions, test our systems and strengthen safeguards."
The politics behind the handshake
The deal also carries a policy argument. OpenAI says it has called for the United States to work with other countries on international technical standards for frontier AI — shared ways of evaluating capabilities, assessing risks and testing safeguards. Without national partners like NASK, OpenAI argues, "global AI safety standards may remain detached from the realities in which AI is actually used."
The context is a contested fight over who gets to test frontier models before release. In September, US officials reportedly asked labs to hold new models back from the UK AI Safety Institute, an episode that ended with the White House reversing the restriction.
OpenAI already works with the US Center for Advancing Standards and Innovation for Super Intelligence (CAISSI) and the UK AI Security Institute. A lab that signs agreements with several national institutes spreads its testing relationships — and signals it is not tied to one government's gatekeeping. The real test will be what emerges in the coming months: whether NASK gets pre-release access, whether it can publish findings OpenAI dislikes, and whether a first joint output appears within a year.
By Hamza Faisal



