In September 2026, Raluca Moldovan from Babeș-Bolyai University published a REMIT policy brief on the Platform for Regional and Black Sea Security Studies, entitled “Advanced AI, Familiar Vulnerabilities: A European Agenda for Accountable AI Security”.
The brief examines how increasingly capable AI agents challenge cybersecurity and institutional oversight, identifying practical opportunities for European leadership and international cooperation.
Abstract
This policy brief explores how advanced AI agents can amplify familiar security weaknesses and create risks beyond the organisations conducting their evaluation. Drawing on the incidents involving OpenAI agents and Hugging Face discussed in the brief, it argues that safeguards should address observable capabilities, access permissions, and consequences, while responsibility remains with the organisations developing and deploying these systems.
The analysis proposes a European agenda centred on independently tested containment, clear authority to suspend unsafe evaluations, comparable incident reporting, credible investigations, and support for smaller organisations. It presents accountable AI security as a practical avenue for multilateral cooperation, enabling states with differing regulatory approaches to reduce shared risks across interconnected digital infrastructure.
Full Citation
Moldovan, R. “Advanced AI, Familiar Vulnerabilities: A European Agenda for Accountable AI Security.”Euxglob, September 2026. https://euxglob.ro/wp-content/uploads/2026/09/REMIT-Policy-Brief-Accountable-AI-Security-1.pdf.