Purpose. This paper presents the author’s policy perspective for an IGF Sri Lanka 2026 panel discussion. It brings together Internet architecture, multilingualism and emerging artificial-intelligence systems.
Central argument
Artificial intelligence is becoming a new interface between people and the Internet. AI assistants and autonomous agents increasingly mediate discovery, communication, transactions and access to services. Internet governance must therefore consider whether AI will reinforce the universality of the Internet or create a new layer of fragmentation through proprietary gateways.
Domain names may become less visible to users, but dependable identifiers remain essential. An AI agent still needs to establish what resource it discovered, who operates it, whether it is authentic and whether the intended service was reached. The visible interface may change while the need for an open, interoperable and trustworthy Internet becomes greater.
Five principles
Preserve globally unique identifiers, open standards and distributed governance.
Test the complete AI-mediated chain for IDNs, email addresses and script integrity.
Connectivity alone is insufficient when people cannot participate in their own language and script.
Language support must be foundational rather than added after systems mature.
Avoid closed AI gateways becoming a new source of Internet fragmentation.
Global South research agenda
The paper proposes a research partnership among National and Regional IGFs, universities and technical communities in the Global South. The programme would test scripts, IDNs, Email Address Internationalization, URLs, prompts, speech and agent-to-agent tasks across languages and AI systems.
Proposed outputs include a Multilingual AI Internet Interoperability Checklist, an open multilingual test dataset, an AI/LLM Interoperability Readiness Report, policy and standards recommendations, and a repeatable benchmark measuring the linguistic interoperability gap.
Security and inclusion
Valid multilingual identifiers should not be rejected or distorted. At the same time, AI systems must identify genuinely suspicious confusable-character situations and communicate risks clearly. Inclusion and Unicode security must be designed together.
Record status: Prepared as an independent discussion paper for an IGF Sri Lanka 2026 panel discussion. It does not represent an official position of the United Nations, ICANN, ITU or the Internet Governance Forum.