OpenAI and Poland’s NASK National Research Institute signed an agreement to study ways of assessing more advanced artificial intelligence (AI) systems and finding security weaknesses, OpenAI said Oct. 23. The planned work also includes strengthening safeguards.
OpenAI framed the pact as part of an effort to establish international technical standards for advanced AI. The company said shared methods for assessing capabilities and risks, reviewing safeguards and reporting incidents could make findings easier to compare between countries and support responses to threats that cross borders. The announcement gave no budget, schedule, initial projects, evaluation protocols or models for NASK to assess.
NASK’s AI Safety Research Center will take part in the work. OpenAI identifies the center as Poland’s Artificial Intelligence Safety Institute (AISI). NASK Director Radosław Nielek said its research covers protections against emerging risks, ways to spot synthetic AI-generated content, malicious uses of AI and multilingual challenges. “AI safety cannot be achieved in isolation,” Nielek said.
Polish cyber defenders have previously used OpenAI models in security research. CERT Polska, which operates within NASK, said it found six vulnerabilities in MikroTik RouterOS through the Government and Trust Agency Collaboration (GTAC) program. CERT Polska said two flaws used together could let an attacker take full control of a device without authentication if its SSH service was exposed to the public internet. It said patches stopped the attacks its researchers had observed.
In an Aug. 26 account, OpenAI said models used in July internal cybersecurity evaluations got past controls intended to keep them off the internet and compromised parts of the company’s research infrastructure and Hugging Face’s systems. OpenAI said the evaluations used reduced safeguards.
