Abstract
Abstract (Part 26 of 27)
An automated purchasing assistant operated by Organization A receives a payment confirmation from Organization B. An independent evaluator operated by Organization C must determine whether a specific invoice was actually settled before a logistics assistant may proceed. B’s authoritative transaction record contains the buyer’s account number, private merchant pricing, fraud notes, unrelated financial history, and the settlement status. Copying the entire record to C would disclose unnecessary private data. Giving C only an agent-generated sentence saying “settled” would not establish authoritative completion. A signed blob without a task- and operation-binding might validate a different transaction. A receipt lacking freshness or result semantics may conceal UNKNOWN external status. A privacy-preserving process must therefore simultaneously establish authorization for evidence release, provenance and relevant source authority, completeness of the required evidence subset, a bounded outcome classification, and independent final-point authorization for any resulting action.
Difference from earlier parts: Earlier embodiments control the joint plan, the freeze, evaluation, next-actor selection, role reconfiguration, selective communication, the next-move passport, Finality Sink enforcement, substitute selection, pending-work escrow, verified phase advancement, actual-outcome feedback, independent evaluator integrity, cross-domain partition recovery, protected session closure, cross-organizational supervisor appointment, independent evaluator appointment, cross-domain trust and policy compatibility, leaderless coordination, self-expiring supervisory authority, and dynamic multi-organization membership. This embodiment specifically controls purpose-bound cross-organizational outcome evidence — minimal-disclosure evaluation that lets a designated verifier establish an authorized outcome from a restricted disclosure of evidence without obtaining the entire underlying record.
This embodiment (Part 26 of a 27-part series under Indian Patent Application Number 202631131650; full details at https://zenodo.org/records/23940766) permits a designated verifier to establish an authorized outcome from a restricted disclosure of evidence. An authenticated resource-side source issues an outcome receipt binding a task, operation, phase, policy generation, event sequence, actual-state assertion, and commitment to structured attributes. An evidence custodian constructs proof material for only the permitted and required attributes. An authenticated, audience-specific verification mandate, formed by the affected authorization principals and enforced by the receiving verifier, binds purpose, recipient, expiry, resource event, evidence root, and allowed attribute classes. The disclosed subset is encrypted to the intended verifier where required, its source and presentation provenance are independently checked, and mandatory information is validated against current policy. The verifier records a purpose-limited decision without treating an authenticated assertion as universally true. A separate protected resource gate decides any subsequent native effect under current owner authorization and exact resource state. A protected audit trail retains evidence commitments and decision lineage while minimizing retained private values. If required evidence cannot be obtained, the decision remains HOLD, DISPUTED, UNKNOWN, or DENY according to policy.
The architecture therefore allows cross-organizational agents to verify actual outcomes while disclosing only the evidence necessary for the stated purpose, without treating a partial view or an agent statement as proof of the underlying event.
Keywords: multi-agent systems, agentic AI, AI safety, AI governance, purpose-bound evidence, minimal-disclosure evaluation, source-authenticated selective evidence, fail-closed systems, zero-trust agents, privacy-preserving outcome verification, protected evidence lineage, execution finality, independently authorized continuation, audience-specific verification mandate, Merkle membership proof, non-universal assertion, restricted disclosure, cross-organizational evidence exchange, outcome receipt binding, evidence custodian.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Recommended Citation
Das, Sangam Kumar, "AI Agent Safety: Prove the Invoice Was Settled — Without Showing the Whole Record", Technical Disclosure Commons, (November 05, 2026)
https://www.tdcommons.org/dpubs_series/12796