New paper: Agentic web blocked by outdated laws, needs norms

The agentic web, where users delegate tasks to AI agents that interact with online platforms on their behalf, is now technically feasible. However, a new arXiv paper (https://arxiv.org/abs/2606.10711) by Cameron Pattison, Matthew Boulos, Noam Kolt, Changbai Li, Tiziano Piccardi, and Seth Lazar reveals a critical bottleneck: outdated laws, terms of service, and informal practices allow platforms to block or degrade agent access, often in secret. No distinction is currently made between 'malicious bots' and AI agents acting with a user's express delegated authority. This blanket blocking prevents consumers and businesses from realizing the full benefits of agentic technologies—such as automated shopping, travel booking, or data aggregation—while leaving platforms vulnerable to abuse.

The paper advocates for building 'normative infrastructure' to accompany the technical protocols and interfaces already emerging. By normative infrastructure, the authors mean a broadly accepted set of laws, norms, and practices that govern agentic access to online properties. They propose principles that enable users' appropriately delegated agents to act online with minimal curbs, only limited as necessary by other legitimate interests. The paper explicitly calls for a society-wide conversation to develop this framework, arguing that piecemeal technical solutions (like better bot detection) are insufficient. Without such normative foundations, the promise of the agentic web—increased efficiency, personalization, and user autonomy—will remain blocked by outdated paradigms that conflate delegation with intrusion.