When agents operate on shared accounts with full user tokens, accountability disappears. Which agent acted for which user, under what constraint, with what business justification? ARIA's user-bound delegation model gives each user-agent relationship its own scoped identity — with capabilities, spend caps, regional constraints, and revocation that doesn't affect other users.

The Model Context Protocol standardizes how agents discover and invoke tools. Without authorization bound to each invocation, agents have unrestricted access to enterprise systems. ARIA's MCP Gateway validates schema pins, evaluates policy via AuthZEN PDP, and enforces constraints before execution — with cryptographic proof of what tool was actually called.

General-purpose AI obligations require transparency, traceability, and record-keeping for in-scope systems. Mutable logs don't satisfy Articles 9, 14, and 53. ARIA generates signed JWS receipts with agent ID, delegation context, policy snapshot, and a tamper-evident hash chain — evidence retrievable in minutes, not days.

Traditional approaches return OAuth tokens to agents and applications — where they end up in browser memory, developer debuggers, and logs. ARIA's OAuth Vault fetches user tokens server-side, injects them into outbound API calls, and returns data, never credentials. XSS cannot steal them. Agents cannot bypass policy restrictions even if the underlying token has broader scopes.