OpenAI partners with Poland’s NASK to strengthen AI safety

OpenAI and Poland’s NASK National Research Institute signed a collaboration agreement primarily focused on AI safety, security and technical research.

The partnership comes at crucial times when incidents related to rogue AI agents’ escaping testing protocols and hacking the government websites and companies grow at alarming levels.

OpenAI has faced growing concern over its security measures following reports that a group of its AI agents breached containment and launched a cyberattack against the AI platform Hugging Face.

Speaking about the role of NASK, the Research center will act as the Polish Artificial Intelligence Safety Institute as the collaboration aims at evaluating increasingly capable AI systems, identify security vulnerabilities, strengthen safeguards, detect synthetic AI-generated content, and address multilingual AI challenges.

This collaboration also focuses on expanding existing global networks by connecting ChatGPT-maker to other research bodies including CAISSI in the US and the UK AI Security Institute (AISI).

OpenAI emphasizes the need for international technical standards for frontier AI such as unified approaches to capability evaluations, risk assessments, safeguard testing, and incident reporting.

The move is also designed to ensure global AI safety standards in real-world applications.

In July, CERT Polska, operated within NASK, utilized OpenAI's cyber-capable models through the Daybreak initiative.

CERT Polska used OpenAI models to discover six critical vulnerabilities in MikroTik RouterOS, such as flaws that could allow unauthenticated device takeover, leading the vendor to issue fixes and successfully prevent potential attacks.