Designing for AI Agents: Identity, Reputation, and Accountability in an Intermediated Web
This article explores the architectural challenges posed by the rise of AI Agents acting on behalf of users, fundamentally changing the interaction model between websites and human users. It highlights the loss of direct user information, which impacts security, fraud prevention, and monetization. The post proposes new mechanisms and protocols focusing on identity, reputation, and accountability to maintain the crucial relationship between sites and users in an agent-driven web.
Read original on Cloudflare BlogThe Rise of AI Agents and Intermediation
The proliferation of AI Agents introduces a significant shift in how users interact with web services. Traditionally, sites directly observed human users, allowing for established mechanisms of identity, reputation, and accountability. AI Agents, operating as intermediaries, abstract away the human user, making it challenging for sites to distinguish between individual users, prevent abuse, or even manage capacity effectively. This intermediation breaks the direct relationship that many web services are built upon.
The Intermediation Principle and Core Challenges
The Intermediation Principle
The business relationship between Users and Sites, and each participant ’s security responsibilities, must be upheld regardless of intermediation. Sites need new tools to recognize, serve, and hold accountable Users acting through Agents, just as they would if the User visited directly.
The core problem is the loss of direct user visibility for site operators. This lack of visibility complicates critical functions like security, fraud prevention, abuse detection, and personalized user experience. When an agent acts for multiple users, or performs actions at a scale a human never could, sites face increased costs and risks without clear accountability. Existing solutions like CAPTCHAs or tightly scoped access controls often fall short because they either reintroduce friction that agents are designed to eliminate or don't address the underlying issue of user identity and accountability.
Architectural Solutions: Protocols, Identities, and Claims
To address the challenges posed by AI Agents, the article suggests developing new protocols and leveraging existing standards for identity and claims. These mechanisms aim to restore the ability for sites to understand who is behind an action, establish trust, and ensure accountability, even when an intermediary agent is involved. This moves away from implicit identity inferred from device fingerprinting towards explicit, verifiable information.
- Protocols: Introduce explicit communication protocols between sites and agents for behaviors, allowed actions, and policy expression. Examples include MOLE for privacy-preserving rate limiting and x402 for web payments.
- Identities: Establish mechanisms for agents to sign requests (e.g., Web Bot Auth - WBA) to identify themselves and their operators. More granular user identity options range from anonymous ('Nothing') to session-based, account-based, or unique per-user identifiers (e.g., government IDs, Proof of Human) for high-risk actions. The goal is data minimization while providing necessary trust signals.
- Claims: Enable sites to request verifiable facts about the real person (e.g., age, cardholder status) via mechanisms that allow operators or third parties to attest to these claims without fully exposing the user's identity.
The proposed solutions require a collaborative ecosystem effort to integrate these new standards and protocols into both site infrastructure and agent platforms. This ensures that the benefits of AI agents (e.g., increased access, automation) can be realized without compromising the security, integrity, and economic models of web services.