Skip to content
INTERNET INFRASTRUCTURE & NETWORKS

Cloudflare Launches Trusted Agent Delegation Coalition to Tackle the Rise of AI Intermediary Traffic

The explosive growth of personal artificial intelligence agents is fundamentally altering the landscape of the internet, creating an unprecedented crisis of visibility and accountability for website operators worldwide. As millions of users begin delegating everyday web tasks to cloud-hosted AI assistants—ranging from automated shopping to complex data retrieval—websites are increasingly struggling to recognize, serve, and hold accountable the actual human beings behind these automated sessions. In response to this shifting paradigm, infrastructure and security leaders are pushing for a unified ecosystem approach, highlighted by the newly formed Trusted Agent Delegation coalition.

Personal AI agents, such as consumer offerings and managed services developed by major tech players like OpenAI and Meta, act as software automations that take instructions from human users and pursue goals by controlling cloud computers or remote browsers. Operating around the clock without human fatigue and executing tasks in parallel, these agents can theoretically perform any action online that a human user could, but at a scale and speed that legacy web architecture was never designed to handle. For decades, the foundational economics and capacity planning of the web have assumed strict human limitations. The rise of agentic traffic threatens to eclipse traditional human browsing entirely, forcing a systemic reckoning for site operators.

The Intermediary Problem and the Cost of Lost Visibility

At their core, remotely hosted AI agents function as digital intermediaries, standing directly between the user and the website. While intermediaries like proxy services and virtual private networks have existed for years, the trajectory of AI agents presents a distinct challenge because traditional methods of managing unwanted traffic are failing. Historically, website operators could simply block intermediary traffic or require users to disable proxy services. However, when a significant portion of incoming traffic originates from AI agents, blanket blocking becomes commercially unviable.

Without direct visibility into who is operating the agent, website operators face a cascading series of operational and financial liabilities. When something goes wrong—whether through accidental user error, policy circumvention, or malicious abuse—the website typically absorbs the primary cost.

This dynamic manifests across numerous operational challenges. In scenarios involving account sharing, a site sees a single automated client representing what might actually be dozens of distinct users sharing a single paid subscription. Similarly, regulatory compliance regarding age and geographic restrictions becomes nearly impossible to enforce when an agent acts as a opaque proxy. Platforms also face increased friction from friendly fraud, where a user authorizes an agent to make a purchase but later disputes the charge, leaving the site unable to prove the transaction originated from the user’s personal device or IP address.

Furthermore, traditional mechanisms for maintaining security and trust are breaking down. Standard human-verification challenges, such as CAPTCHAs, are fundamentally misaligned with automated agents that are never expected to be human in the first place, serving only to add frustrating friction without providing meaningful security signals. Risk-based authentication systems, which rely on consistent user behavior, device fingerprints, and geographic locations, fail when agents consistently route requests through dynamic cloud infrastructure. Most critically, traffic commingling prevents websites from effectively moderating abuse; when a shared agent is exploited for credential stuffing or scraping, blocking the agent inadvertently locks out all the legitimate users relying on that same service.

Re-establishing Identity, Reputation, and Accountability

Addressing these systemic vulnerabilities requires returning to the foundational pillars of internet security and fraud prevention: identity, reputation, and accountability. Historically, these components allowed website operators to answer three critical operational questions: Is the entity acting being truthful about who they are? Is that entity trustworthy? And who ultimately answers for the action taken?

Agents aren’t humans (but they kinda are)

In the current ecosystem, however, these questions go unanswered. Industry observers note that while solutions like human-in-the-loop workflows—where an agent pauses to pass a CAPTCHA back to the human user—may help mitigate occasional errors caused by software misalignment, they do little to solve broader architectural issues like traffic commingling or long-term accountability. Similarly, tightly scoped access permissions tell a website what an agent was technically allowed to do, but they fail to establish who takes responsibility when the agent makes a mistake or executes an unauthorized transaction.

To bridge this gap, technologists are advocating for open, interoperable standards and protocols that allow websites and agents to collaborate seamlessly. Existing frameworks are already being adapted to address these needs. For instance, emerging protocols like MOLE aim to facilitate rate limiting and access control without relying on invasive, user-specific identifiers. Meanwhile, Web Bot Auth provides a secure cryptographic method for agents to sign their requests, allowing websites to verify the identity of the agent and its underlying operator.

Additional standardization efforts are focusing on how identity exists on a graduated spectrum. By leveraging privacy-preserving mechanisms such as Privacy Pass and cryptographic tokens, websites can establish trust and verify eligibility without necessarily demanding full personal identification, thereby respecting data minimization principles. Where more sensitive actions are required, such as age verification or cardholder authentication, interoperable claims mechanisms can securely pass verified facts from operators or trusted third parties directly to the destination site.

Towards Open Standards and Ecosystem Collaboration

Real-world solutions to the challenges posed by AI agents cannot be solved through fragmented, vendor-specific mechanisms or isolated proprietary blocks. Industry leaders emphasize that the transition to an agent-driven web requires open standards rooted in practical operational experience. This realization has spurred cross-industry collaboration among major AI operators, identity verification companies, payment providers, merchants, publishers, proxy operators, and cloud infrastructure firms to align operational requirements with ongoing work in standards organizations like the Internet Engineering Task Force.

To formalize these efforts and drive tangible progress, stakeholders have announced the formation of the Trusted Agent Delegation coalition. The coalition is designed to bring together diverse representatives across the internet ecosystem to examine existing protocols, analyze systemic challenges, and formulate well-defined technical proposals for relevant IETF working groups. By bridging the gap between running code and emerging specifications across initiatives such as Web Bot Auth, OAuth, and privacy-preserving protocols, the coalition aims to establish a sustainable foundation for the future of the web.

In the interim, platform operators are beginning to deploy immediate, production-ready tools to help website owners regain visibility and control over automated traffic. Modern security analytics dashboards are increasingly categorizing personal assistants as distinct traffic segments, enabling administrators to monitor request volumes, path traversals, and error rates associated with specific AI agents. Advanced configuration policies allow site operators to set granular access rules on a per-agent basis, welcoming value-driving automation while restricting or blocking unwanted crawlers without disrupting broader search engine optimization or training policies.

Furthermore, early-access account abuse protections are beginning to connect agent traffic back to individual accounts, allowing security teams to mitigate compromised sessions via hashed user identifiers rather than executing blunt blocks that inadvertently penalize entire cohorts of legitimate users. As these technological building blocks mature and collaborative standards take root, the broader digital ecosystem continues to search for a sustainable balance—one that preserves the immense utility and accessibility of personal AI agents while safeguarding the fundamental business relationships and security responsibilities that sustain the web.