Policy

OpenAI and Poland's NASK forge AI safety research partnership

OpenAI expands its international AI safety collaboration network by partnering with Poland's NASK National Research Institute to advance safeguards and security standards.

Last verified:

OpenAI expands AI safety institutes network to Eastern Europe

According to the OpenAI Blog, OpenAI and Poland’s NASK National Research Institute have formalized a partnership to collaborate on AI safety research, frontier AI evaluation, and security safeguards development. The agreement positions NASK’s AI Safety Research Center—functioning as Poland’s Artificial Intelligence Safety Institute (AISI)—as a key node in OpenAI’s international safety infrastructure alongside the US-based Center for Advancing Standards and Innovation for Super Intelligence (CAISSI) and the UK AI Security Institute (AISI).

Research focus areas and vulnerability assessment

The collaboration will concentrate on evaluating increasingly capable AI systems, identifying potential security vulnerabilities, and strengthening technical safeguards. According to OpenAI, this work builds on existing momentum: CERT Polska, which operates within NASK and previously gained access to OpenAI’s cyber-capable models through Daybreak in July 2026, used that capability to uncover six vulnerabilities in MikroTik RouterOS. The vendor released patches in response to CERT Polska’s findings, which the research team confirmed prevented the attacks they had observed.

Rationale for national embedding of AI safety work

OpenAI positions this partnership as essential to grounding international AI safety standards in real-world deployment contexts. According to the OpenAI Blog, without national partners embedded in their own regulatory and operational ecosystems, “global AI safety standards may remain detached from the realities in which AI is actually used.” NASK brings specialized expertise in protecting public services and Poland’s digital information systems—context that shapes both the research questions and the practical applicability of safety findings.

The partnership also reflects OpenAI’s broader call for international technical coordination. OpenAI has advocated for harmonized approaches to evaluating AI capabilities, assessing risks, testing safeguards, and reporting incidents, enabling countries to compare evidence and respond to cross-border risks effectively.

Why This Matters

This partnership signals a consolidation strategy in AI safety governance: as AI systems become more capable and globally deployed, safety institutes localized within national contexts—rather than centralized in Silicon Valley—become more credible and effective at generating standards that countries will actually adopt. For organizations building AI safety practices, the precedent of NASK-OpenAI collaboration suggests that vendor-led safety work gains legitimacy and enforcement potential when embedded within government research institutions. The vulnerability discovery workflow with CERT Polska also demonstrates that cyber-capable models themselves can be repurposed to identify AI system weaknesses, creating a feedback loop for safety improvement. Teams evaluating AI vendors should monitor whether similar partnerships emerge in other regions—the absence of a domestic safety institute in a given country may indicate slower adoption of frontier AI safety practices there.

Frequently Asked Questions

What is NASK and what role does it play in this partnership?

NASK is Poland's National Research Institute, operating the country's AI Safety Research Center and serving as the Polish Artificial Intelligence Safety Institute (AISI). The institute brings expertise in protecting public services and Poland's digital infrastructure to the collaboration.

How does this partnership connect to OpenAI's broader safety work?

This agreement extends OpenAI's existing safety collaborations with organizations like the US CAISSI and UK AI Security Institute, positioning national research institutions as anchors for developing internationally coherent AI safety standards.

What practical security work has already emerged from OpenAI's Poland collaboration?

CERT Polska, which operates within NASK, used OpenAI's cyber-capable models to discover six vulnerabilities in MikroTik RouterOS in July 2026, leading to vendor fixes that prevented observed attacks.

#AI safety #international collaboration #cybersecurity #Poland #research institutes