Poland's NASK gains access to OpenAI's restricted GPT-5.5 Cyber model

Poland's NASK, the state research institute, has been given access to OpenAI's GPT-5.5 Cyber, according to reporting on 2 July 2026 that began with an interview in Rzeczpospolita. CERT Polska, the response team inside NASK, is the group that will use it, and NASK's director says the institute is already doing so successfully.

Access comes through OpenAI's Trusted Access for Cyber programme for government bodies, part of an initiative called Daybreak, which aims to shift the balance of power towards defenders. The model is a specialised system with fewer restrictions than public subscription models, and it is available only to vetted organisations. NASK received it free of charge. Rzeczpospolita and Money.pl report Poland as the only country in its region with access and one of a handful worldwide, alongside G7 states, although one Polish headline calls it the third in Europe.

NASK uses it for malware analysis, forensics and vulnerability hunting. Through an API in an agent setup that produces reports, it scans public administration and company systems as well as NASK's own tools.

Director Radosław Nielek credits the institute's international reputation for the access. He calls the model a force multiplier, not a replacement for experts, and a large qualitative step without being a full game changer. He puts it on a par with Anthropic's Mythos. Money.pl reads the grant as approved by the US administration, and Nielek says OpenAI must tell Washington who receives the tool.

The setup matters as much as the model. Rather than a chat window, NASK runs it through an API inside an agent arrangement, where the model works through targets and writes up what it finds as reports for the human analysts. That lets a small team cover far more systems than it could by hand, which is the sense in which Nielek calls it a multiplier. He is careful to say that it supports experts instead of replacing them.

The access has conditions and a political edge. Trusted Access for Cyber is limited to vetted government bodies, and the less restricted behaviour that makes the model useful for analysing malware is exactly why it is not sold to the public. Nielek says NASK has nothing to announce with Anthropic, though the institute remains open to it, and Money.pl presents the grant as something Washington had to sign off.

Poland's position is described differently by different outlets. Rzeczpospolita and Money.pl say it is the only country in its region with access and one of very few worldwide, while a Centrum XP headline, which was not read in full, calls it the third in Europe. The reports do not settle the ranking, and the common point is simply that Poland sits in a small group that includes G7 states.

The security context is part of the framing. Money.pl places the access against rising threats from Russia, and it quotes a Fortinet 2026 report counting more than 67 billion brute-force attempts over the past year. For a state institute that scans public administration and company systems, those numbers show why faster analysis is wanted.

Daybreak's stated aim, in both Rzeczpospolita and Money.pl, is to reverse the power asymmetry that favours attackers, who need to find only one weakness while defenders must close all of them. NASK's director ties the institute's selection to years of research and its name abroad in technology circles.

CERT Polska head Marcin Dude says experts need such tools to find and fix flaws before attackers do. Money.pl sets the story against growing threats from Russia and cites a Fortinet report counting more than 67 billion brute-force attempts in a year. NASK plans to act as a partner CERT for the region, sharing findings on vulnerabilities without naming the model or method, and will not pass the model on.