{"page":"/blog/ai-agents-data-leakage","modified":"2026-11-10","license":"https://creativecommons.org/licenses/by/4.0/","columns":["Data class","Vendor-hosted agent, vendor's model","Your agent, external model API","Your agent, model in your environment","Keep away from agents"],"rows":[["Public inventory and rate cards","Yes","Yes","Yes","No need"],["Aggregated performance reports","Yes, if the DPA bars training and sets retention","Yes, if the DPA bars training and sets retention","Yes","No need"],["Deal terms and negotiated prices","Yes, if the vendor is your contractual counterparty for that deal","Yes, if the provider's terms bar training","Yes","Consider for strategic deals"],["Bid-level and log-level data","No","Only aggregates derived inside your environment","Yes, with outbound allowlist","Raw logs with device IDs"],["Clean-room outputs","Yes, inside the clean room's own tools","Yes, if outputs pass aggregation thresholds","Yes","Row-level match tables"],["Audiences and hashed identifiers","No","No","Yes, if a lawful basis covers the processing","Yes, unless a privacy-preserving match is used"],["Unreleased briefs and creative","Yes, if confidentiality terms cover the vendor","Yes, if the provider's terms bar training","Yes","Highly sensitive launches"],["Credentials","Never in prompts","Never in prompts","Never in prompts","Always in a secret store"]],"sourceUrls":[[],[],[],[],[],[],[],[]]}