The internet is facing a radical shift in the nature of traffic; while interaction used to occur directly between the user and the website, AI Agents have begun to emerge as digital intermediaries performing tasks on behalf of millions of humans. This shift, despite its efficiency, creates a deep technical and security gap, as websites see the "agent" instead of the "real user," leading to a loss of ability to identify identity, prevent fraud, and ensure accountability.
The Erosion of the Site-User Relationship
Current web business and security models rely on the ability of websites to recognize users and track their behavior. With the entry of agents (such as Meta's Muse or OpenAI's dots), the agent becomes an "intermediary" that obscures essential data. This ambiguity imposes complex technical challenges, including:
- Friendly Fraud: Where a user authorizes an agent to make a purchase, then later disputes the transaction claiming it was unauthorized, and websites find it difficult to prove otherwise because the data comes from the agent's cloud infrastructure rather than the user's device.
- Traffic Commingling: When a single user abuses a shared agent, websites may be forced to block the agent entirely, leading to the blocking of thousands of legitimate users associated with it.
- Resource Exhaustion: The ability of agents to work around the clock and at speeds exceeding humans leads to massive pressure on expensive endpoints such as search engines and chat services.
- Collapse of Identity Verification: Tools like CAPTCHA are no longer effective because agents are not human, and Know Your Customer (KYC) processes assume a real person is behind the screen, which is not verified when using an agent.
Why Does This Development Matter Technically?
The problem does not lie in the existence of automation, but in The Intermediation Principle; the commercial relationship and security responsibilities between the user and the website must remain intact even in the presence of an intermediary. Without new protocols, websites will be forced to either block agents entirely (reducing user access to their services) or accept uncalculated security and economic risks.
Towards New Collaborative Protocols
Cloudflare proposes a technical vision based on building open standards that allow agents and websites to "collaborate" rather than collide. This vision includes several technical paths:
- Web Bot Auth (WBA): A protocol that allows an agent to digitally sign its requests, providing a secure way to identify the platform and the entity operating the agent.
- Reputation Management (MOLE): A protocol aimed at allowing rate limiting based on user reputation without the need to reveal their personal identity, thus preserving privacy.
- TAD (Trusted Agent Delegation) Coalition: Cloudflare announced the launch of this coalition to bring together representatives from various sectors (operators, payment providers, publishers) to set global standards to address these gaps at the IETF.
What Changes Practically for Cloudflare Users?
Cloudflare has already begun providing concrete tools to deal with this new reality, including classifying "personal assistant" traffic in security analytics, and providing control policies that allow websites to permit some agents and block others based on specific pages, in addition to the feature of linking the agent to the user (via Account Abuse Protection) to ensure that an abusive user can be penalized without blocking the agent for the rest of the users.