{"page":"/blog/ai-agents-data-leakage","modified":"2026-11-10","license":"https://creativecommons.org/licenses/by/4.0/","columns":["Topic","MCP (2026-07-28)","A2A (v1.0)","AdCP (3.2.x)","Left to you"],"rows":[["Transport security","Covered in authorization security section","HTTPS or TLS must be used in production","Bearer, mutual TLS or signed requests","Certificate and key management"],["Token scope","Audience-bound tokens; no passthrough","Credentials obtained out of band; every request authenticated","Signing optional in 3.x","Short token lifetimes; one token per resource"],["User consent","Hosts must obtain consent before exposing user data","Not addressed in the protocol","No consent signal carried (no TCF or GPP)","Consent capture and checks"],["Authorisation rules","Server's responsibility","\"Implementation-specific\"","Governance checks as a hook","Per-tool, per-field permissions"],["Personal data in fields","Not addressed","Not addressed","Hashed identifiers reject cleartext; free text not scanned","Redaction and scanning"],["Logs and retention","Not addressed","Agents may persist task history","Audit logs immutable; retention is the deployer's job","Retention schedule, deletion"],["Model training","Not addressed","Not addressed","DPA must state training use","Contract terms"]],"sourceUrls":[[],[],[],[],[],[],[]]}