OpenAI and Poland’s NASK partner to research AI safety and security

Article Summary

OpenAI and Poland’s NASK National Research Institute have signed an agreement to collaborate on AI safety research, security safeguards, and risk assessment. This partnership aims to integrate national expertise into global safety standards and technical research. Previously, CERT Polska, which operates under NASK, utilized OpenAI’s cyber-capable models to identify six vulnerabilities in MikroTik RouterOS.

AI Analysis

## Core Assessment

OpenAI’s agreement to explore AI safety and security research with Poland’s NASK National Research Institute follows a familiar playbook: expand the vendor's network of national safety institutes beyond the US and UK to build buy-in for international frontier AI standards. What keeps this from being purely ceremonial diplomacy is the operational precedent behind it. Poland’s national cyber defense arm, CERT Polska (operating inside NASK), previously used OpenAI’s cyber-capable models through the Daybreak program to uncover six MikroTik RouterOS vulnerabilities—flaws severe enough to allow unauthenticated device takeovers before patches were verified.

The technical trade-off lies in granting state-linked research bodies early access to offensive and defensive frontier tools. While CERT Polska demonstrated a clean defensive outcome by reporting flaws and confirming vendor patches, formalizing these pipelines tests the boundary between safety evaluation and dual-use capability proliferation. NASK's AI Safety Research Center gets a seat at the table to shape capability evals and incident reporting, but the actual framework remains an exploratory memorandum; concrete governance metrics and testing protocols have yet to be delivered.

## Impact and Advice

* **Enterprise Security Operations Leads**: Treat model-assisted vulnerability research as an urgent forcing function for edge-device patching. Since teams like CERT Polska proved frontier models can discover chained zero-day exploits in widely deployed network appliances like MikroTik RouterOS, assume automated offensive auditing will find your unpatched network perimeters first.

## Bottom Line

Signing an exploratory safety pact costs nothing; the real signal is that frontier models are already finding critical network zero-days in the wild.

Explore this story further