OpenAI and Poland’s NASK Nationwide Analysis Institute have signed an settlement to discover collaboration on AI security, safety and technical analysis, OpenAI introduced on October 23, 2026.
OpenAI mentioned the partnership will deliver collectively experience from each organizations to higher perceive rising dangers, strengthen safeguards, and discover how superior AI can contribute to a safer and resilient digital ecosystem. Below the settlement, NASK’s AI Security Analysis Heart, which serves because the Polish Synthetic Intelligence Security Institute, and OpenAI will collaborate on AI security analysis, exploring methods to guage more and more succesful AI methods, establish potential vulnerabilities and strengthen safeguards.
OpenAI mentioned its present work with the US Heart for Advancing Requirements and Innovation for Tremendous Intelligence, the UK AI Safety Institute and different public analysis establishments has already helped strengthen safeguards, establish and repair safety vulnerabilities in AI brokers, and advance frontier AI security.
Statements From NASK and OpenAI Management
NASK Director Radosław Nielek mentioned collaboration with OpenAI is a pure step for the institute and its AI Security Analysis Heart. He mentioned the middle’s work spans growing safeguards towards rising dangers, detecting artificial AI-generated content material, and understanding each malicious makes use of of AI and the multilingual facets of those challenges. AI security can’t be achieved in isolation, he mentioned, and solely cooperation between public AI security institutes and frontier AI builders could make AI actually dependable and protected to make use of. By combining NASK’s analysis experience with OpenAI’s technical capabilities, he mentioned, the 2 organizations can higher anticipate dangers, strengthen safeguards and construct public belief within the applied sciences.
Tom Duff Gordon, vp and head of EMEA coverage at OpenAI, mentioned Poland is dwelling to distinctive scientists and technical expertise, a lot of whom have helped lead OpenAI’s personal analysis. He mentioned that as AI turns into extra succesful it’s vital to attract on experience past OpenAI’s personal group to problem its assumptions, take a look at its methods and strengthen safeguards, and that the settlement with NASK opens the door to doing that collectively whereas bringing Poland’s analysis experience into the worldwide effort to make AI safer.
A Wider Push for Worldwide AI Requirements
OpenAI mentioned the NASK collaboration displays the worth of bringing nationwide experience into its wider name for worldwide technical requirements for frontier AI. The corporate has urged america to work with nations worldwide on such requirements, drawing on AI security institutes and requirements our bodies, and mentioned widespread approaches to evaluating capabilities, assessing dangers, testing safeguards and reporting incidents may also help nations examine proof and reply to dangers that cross borders. With out nationwide companions like NASK, OpenAI mentioned, international AI security requirements might stay indifferent from the realities wherein AI is definitely used.
In a September 21, 2026 publish, OpenAI mentioned america ought to lead an effort to develop international technical requirements for frontier AI, together with for recursive self-improvement, a course of wherein AI methods more and more drive growth of successive generations of AI. The publish named three challenges such requirements would handle: fragmentation, the place totally different nations’ evaluations, reporting necessities and incident definitions may battle; collective motion, the place nations appearing independently can produce outcomes no nation needs; and uneven capability, as a result of frontier growth exercise and associated experience aren’t evenly distributed the world over.
The publish pointed to AI security institutes already established in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India and the UK as a community that would facilitate standard-setting.
It proposed requirements for evaluating AI progress related to recursive self-improvement and the quantity of autonomous analysis inside an AI firm; for human oversight of automated AI analysis, together with which automated analysis processes ought to set off quick human evaluate; and for incident classification, monitoring, reporting and response, resembling widespread severity ranges and reporting thresholds. OpenAI mentioned these technical requirements wouldn’t be licenses, obligatory prerelease evaluate or approval necessities for AI fashions, and that nationwide governments would determine whether or not and learn how to incorporate them into their very own authorized methods.
Earlier Cybersecurity Work in Poland
The settlement additionally builds on OpenAI’s present work with cyber defenders in Poland. In July 2026, CERT Polska, which operates inside NASK, gained entry to OpenAI’s cyber-capable fashions by Dawn. OpenAI mentioned the CERT Polska staff used its fashions to assist uncover six vulnerabilities in MikroTik RouterOS, together with flaws that might be mixed to take management of affected units with out authentication. The seller launched fixes, which CERT Polska confirmed prevented the assaults it noticed, based on OpenAI’s account.
NASK’s personal July 2, 2026 announcement mentioned the institute gained entry to GPT-5.5 Cyber, a specialised AI mannequin created by OpenAI for detecting and analyzing cyber threats, by the Authorities Trusted Entry for Cyber program for presidency models. NASK characterised Poland because the third European nation, and one in all few worldwide, to obtain such entry. It described Dawn as an initiative to maintain defenders a step forward of criminals in utilizing AI for cybersecurity, with GPT-5.5 Cyber utilized to malware evaluation, digital forensics and vulnerability discovery to speed up the total cycle of detecting and repairing vulnerabilities.
Nielek mentioned on the time that AI instruments stay instruments that work correctly solely in human fingers, with an individual all the time standing on the finish of the evaluation a mannequin performs and verifying what the device has performed. CERT Polska head Marcin Dudek mentioned the staff’s specialists want entry to such instruments to effectively detect and restore vulnerabilities earlier than attackers do, and added that CERT Polska already shares its information from utilizing giant language fashions on its web site and plans to proceed.
