The United Nations’ digital technology body is moving to establish international guardrails for autonomous artificial intelligence, launching a new initiative on Thursday aimed at ensuring AI agents remain identifiable, trustworthy, and subject to meaningful human control.
The International Telecommunication Union (ITU) announced the creation of a dedicated Focus Group during the AI for Good Global Summit in Geneva. The move comes as a new generation of AI systems—designed to act independently on behalf of users—raises urgent questions about accountability, impersonation risks, and the potential for unauthorized decision-making in critical sectors.
AI agents represent a significant leap beyond chatbots. These systems are engineered to negotiate, transact, and execute complex tasks autonomously, from scheduling meetings and making purchases to managing intricate business processes. While the productivity gains could be enormous, the ITU warns that the technology also opens the door to agents impersonating individuals or making consequential choices without proper oversight.
“AI agents will soon negotiate, transact and make decisions on our behalf,” said Debora Comparin, Co-Chair of the newly formed Focus Group. “Common international foundations are needed to establish who the agents are and how and when they can be trusted.”
The Focus Group will bring together technical, policy, and legal experts to develop frameworks that specifically address high-stakes applications, including financial transactions and the operation of critical infrastructure. The group’s work will kick off with its first meeting in Paris in November, followed by a second session in Geneva in January.
A Showcase of Responsible AI in Finance
The urgency of the ITU’s mission was underscored by real-world deployments presented at the same summit. South Korea’s KakaoBank, a leading digital lender, took the stage at the “Innovate for Impact” session to detail how it is already putting principles of sustainable and trustworthy AI into practice within the financial sector.
KakaoBank presented a methodology that consolidates multiple AI functions—such as information retrieval, document analysis, and anomaly detection—into a single, finance-specific AI model. Rather than operating several large-scale models in parallel, the bank’s unified approach slashes computing resource consumption, graphics processing unit (GPU) memory usage, and power demands, while simultaneously boosting performance on financial document comprehension through domain-specific training.
The presentation, selected as one of the summit’s “Winning AI Use Cases,” demonstrated how a single model can reliably deliver a diverse range of financial services, striking a balance between operational efficiency and sustainability.
The bank’s safety technologies also drew significant attention. Two specific tools were officially listed in the summit’s results report: an AI-based smishing detection service that screens text messages for fraud, and an “AI input guardrail” designed to block prompt injection attacks targeting large language models (LLMs).
“It is deeply meaningful to introduce KakaoBank’s financial AI research achievements as a best practice on a global stage themed around sustainable AI,” a KakaoBank representative said during the event. “We will continue to advance AI technology optimized for the financial environment to deliver tangible value to customers and create safe, convenient financial services.”
The Road to Global Standards
The ITU’s initiative reflects a growing consensus among global regulators and industry players that the breakneck pace of AI development demands proactive, internationally coordinated governance. The Focus Group’s mandate is not to build binding regulations overnight, but to lay the technical and policy groundwork that could eventually underpin formal standards.
Key questions the group will tackle include how to verify the identity of an AI agent acting on a person’s behalf, what constitutes meaningful human oversight in automated decision chains, and how to design systems that cannot be easily weaponized for impersonation or fraud.
The choice of Paris and Geneva as the venues for the group’s initial meetings signals the centrality of European regulatory thinking in this sphere. The European Union has already enacted its AI Act, and the ITU’s work is expected to complement, rather than duplicate, those regional efforts by aiming for a truly global baseline.
For the financial industry, the stakes are particularly high. As banks and fintech firms race to deploy AI agents for everything from customer service to algorithmic trading, the absence of clear international norms creates a patchwork of compliance risks. KakaoBank’s showcase in Geneva illustrates that some institutions are not waiting for regulation to arrive—they are building the technical architecture for safe AI now, betting that early movers will shape the standards that follow.
The ITU’s Focus Group on AI agents represents a critical test of whether multilateral institutions can keep pace with Silicon Valley and Shenzhen. With its first deliverables expected after the January meeting in Geneva, the world will soon see if common international foundations for trustworthy AI agents can move from aspiration to reality.