Eigen RadarAI
Archive
Analysis

NASK and OpenAI sign an agreement to research AI safety

Polish research institute NASK and OpenAI have signed an agreement on AI safety research. The work covers evaluating advanced systems, finding vulnerabilities and testing safeguards, with attention to risks that differ across languages. NASK’s safety research center leads the Polish side, building on earlier cybersecurity cooperation between the organizations.

Artificial Intelligence··Morning
Two closed blue folders and a pen rest on a wooden conference table beneath a Polish flag, with empty chairs behind.

The agreement establishes joint safety research

NASK, Poland’s research institute, and OpenAI have signed an agreement on artificial-intelligence safety research. The agreement brings the institute and model developer together for technical work on the safety of advanced systems.[1], [2]

NASK’s AI Safety Research Center, which serves as Poland’s AI safety institute, leads the Polish participation. The proposed work includes assessing system capabilities, identifying vulnerabilities, testing safeguards and developing approaches to incident reporting. The partners plan to combine research expertise with experience developing models to identify emerging risks earlier.[1]

Language differences enter the risk assessment

NASK director Radosław Nielek described work on emerging threats, detecting synthetic content and malicious use. The language used in a conversation is another research concern because misuse risks can differ across languages. NASK brings experience protecting public services and Poland’s digital information environment. OpenAI’s regional public-policy head Tom Duff Gordon said external experts can challenge the company’s assumptions and help evaluate its systems.[1]

Earlier cybersecurity cooperation supplies the background

Earlier work involved CERT Polska, the cybersecurity team within NASK. It received access to advanced cyber models through Daybreak in July and found six vulnerabilities in MikroTik RouterOS, software used in network devices. Some flaws could be combined to take control without authentication. The vendor released fixes, which CERT Polska confirmed blocked the observed attacks. Those findings precede the newly signed research agreement.[1]

References

  1. News sourceNASKNASK and OpenAI sign an AI safety research agreement↩1↩2↩3↩4
  2. News sourcepb.plOpenAI and NASK agree to research AI safety↩
AI safety and regulationChip wars and the supply chainThe frontier model race

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →