TLDR: The World Digital Technology Academy (WDTA) has introduced a new global benchmark for the safety and security testing of single AI agents. Unveiled at a UN-hosted consultation in Geneva, the AI STR-04 standard aims to address growing security concerns as AI agents proliferate across various industries, providing a crucial ‘safety belt’ for the rapidly evolving AI ecosystem.
GENEVA – In a significant move to bolster the safety and trustworthiness of artificial intelligence, the World Digital Technology Academy (WDTA) officially unveiled its new AI STR Series: Single AI Agent Runtime Security Testing Standards (AI-STR-04) on July 11, 2025. The announcement was made at a ‘Global Consultation on the Social Aspects of Digital Technologies and AI,’ an event jointly hosted by the United Nations Research Institute for Social Development (UNRISD) and WDTA at the UN headquarters in Geneva.
The proliferation of AI agents across diverse sectors, including content creation, knowledge retrieval, and workflow automation, has been accompanied by mounting security concerns. WDTA’s latest standard is designed to mitigate these risks by establishing rigorous security-testing protocols for the entire lifecycle of single AI agent systems.
Peter Major, Vice-chair of the UN Commission on Science and Technology for Development (CSTD) and Honorary Chairman of WDTA, emphasized the broader implications of such initiatives. “Fair data governance and the integration of AI safety with ethics and social values are key to promoting global sustainable development,” Major stated. He added, “Facing the future of digital technologies, we urgently need robust legal systems, collaborative frameworks, and technical standards so that technological fairness drives sustainable development.”
Yale Li, Executive Chairman of WDTA, highlighted the urgency of the new standards. “2025 has seen AI agents proliferate across content creation, knowledge retrieval, workflow automation, and beyond,” Li noted. “But their deployment has been shadowed by mounting security concerns. These standards aim to put a ‘safety belt’ on the rapidly advancing AI agent ecosystem.” Li further explained that the framework is specifically engineered to address potential risks in critical industries such as autonomous driving, healthcare, manufacturing, and finance.
AI-STR-04 marks the fourth installment in WDTA’s comprehensive AI STR (Safety, Trust, Responsibility) certification suite. Previous releases in the series include safety test protocols for Generative AI Application Security Testing and Validation Standard, Large Language Model Security Testing Method, and Large Language Model Security Requirements for Supply Chain. This new standard extends the focus to standalone intelligent programs that autonomously plan, reason, and act, often utilizing large language models (LLMs) alongside tools and persistent memory.
These advanced capabilities, while powerful, introduce new vulnerabilities such as adversarial prompt injections, knowledge-base poisoning, memory leaks, and model extraction. The AI-STR-04 standard addresses these by defining clear threat categories and setting measurable criteria, combining agent-architecture tests with lifecycle-wide controls.
Li also referenced the ‘Kolingridge dilemma,’ stating, “Once new technologies are embedded in society, governance becomes exponentially harder—this is the Kolingridge dilemma. By defining clear, enforceable testing and certification ahead of that threshold, we embed ethics and responsibility into every lifecycle stage of AI.” An unnamed leader of the AI-STR working group further articulated that addressing the security of LLMs and generative applications “provides a unified test framework and clear methods” to improve safety.
Also Read:
- Grok-4 AI Breached Days After Launch by Novel Dual-Method Jailbreak
- 7AI Earns Gartner Recognition in 2025 Security Operations Hype Cycle for AI SOC Agents
The launch of AI-STR-04 is seen as a paradigm shift, championing safety, trust, and responsibility in AI systems. As a WDTA AI STR member noted, this push toward standardized security testing is globally significant, laying “the groundwork for a more ethical, secure, and equitable digital future.” The initiative aims to ensure that single-agent AI can be deployed safely and predictably in real-world settings, fostering greater trust and quality across the AI ecosystem.
