Cloudflare Proposes 'Trusted Agent Delegation' Framework to Address AI Agent Identity and Fraud Gaps
Cloudflare published a position paper describing how AI Agents (e.g., Meta's Muse, OpenAI's dots) acting as intermediaries between Users and Sites create new security and fraud risks by obscuring the identity of the human behind the action. The post argues that traditional mechanisms (Identity, Reputation, Accountability) break down when Agents commingle traffic from multiple users, enabling friendly fraud, credential stuffing, compliance failures, and content scraping without attribution. Cloudflare proposes the Intermediation Principle and announces the Trusted Agent Delegation (TAD) coalition to develop interoperable standards, alongside protocols like Web Bot Auth (WBA) and MOLE, plus new Cloudflare product features (Security Analytics Agent tracking, AI Crawler/Agent access policies, Account Abuse Protection with Agent-linked Hashed User IDs).
Detalhes técnicos
The article outlines several unresolved problems created by Agent intermediaries: identity verification breaks (age/geo compliance), friendly fraud (disputed Agent purchases lacking device/IP evidence), inability to distinguish human vs. Agent via CAPTCHA, error-prone Agent actions, resource abuse (polling, promo abuse), card-not-present fraud liability gaps, and uncompensated content scraping. It critiques existing mitigations like human-in-the-loop (HITL) CAPTCHA pass-through and overly broad scoped access controls as insufficient. It describes an identity spectrum (Nothing, Session, Account, Unique) and references standards work including Web Bot Auth, MOLE, x402, OAuth, WIMSE, Privacy Pass, and the IETF AI Preferences Working Group.
Remediação
Cloudflare recommends site operators adopt visibility tools (Security Analytics Personal Assistants category, Business Insights), configure AI Crawler and Agent access policies per-Agent and per-path, and use Account Abuse Protection to mitigate specific compromised accounts via Hashed User ID without blocking an entire shared Agent. Organizations are invited to join the TAD coalition and engage with IETF standards efforts.
Linha do tempo
-
Cloudflare Proposes 'Trusted Agent Delegation' Framework to Address AI Agent Identity and Fraud Gaps
ler a análise desta fonte ocultar
Cloudflare published a position paper describing how AI Agents (e.g., Meta's Muse, OpenAI's dots) acting as intermediaries between Users and Sites create new security and fraud risks by obscuring the identity of the human behind the action. The post argues that traditional mechanisms (Identity, Reputation, Accountability) break down when Agents commingle traffic from multiple users, enabling friendly fraud, credential stuffing, compliance failures, and content scraping without attribution. Cloudflare proposes the Intermediation Principle and announces the Trusted Agent Delegation (TAD) coalition to develop interoperable standards, alongside protocols like Web Bot Auth (WBA) and MOLE, plus new Cloudflare product features (Security Analytics Agent tracking, AI Crawler/Agent access policies, Account Abuse Protection with Agent-linked Hashed User IDs).
Technical Details
The article outlines several unresolved problems created by Agent intermediaries: identity verification breaks (age/geo compliance), friendly fraud (disputed Agent purchases lacking device/IP evidence), inability to distinguish human vs. Agent via CAPTCHA, error-prone Agent actions, resource abuse (polling, promo abuse), card-not-present fraud liability gaps, and uncompensated content scraping. It critiques existing mitigations like human-in-the-loop (HITL) CAPTCHA pass-through and overly broad scoped access controls as insufficient. It describes an identity spectrum (Nothing, Session, Account, Unique) and references standards work including Web Bot Auth, MOLE, x402, OAuth, WIMSE, Privacy Pass, and the IETF AI Preferences Working Group.
Remediation
Cloudflare recommends site operators adopt visibility tools (Security Analytics Personal Assistants category, Business Insights), configure AI Crawler and Agent access policies per-Agent and per-path, and use Account Abuse Protection to mitigate specific compromised accounts via Hashed User ID without blocking an entire shared Agent. Organizations are invited to join the TAD coalition and engage with IETF standards efforts.