# AuthPros — Comprehensive Public Information for AI Agents (llms-full.txt) > AuthPros — Decisioning for Agentic Transactions. AuthPros develops an advisory trust decisioning approach for agentic commerce, assessing identity, delegated authority, intent, contextual intelligence, and institutional policy. Evidence integrity and production deployment claims require capability-specific verification. --- ## 1. Executive Summary & Brand Identity - **Organization**: AuthPros - **Domain**: https://www.authpros.io - **Canonical Concept**: Know Your Agent (KYA) for Agentic Commerce & Agentic Transactions - **Mission**: As autonomous AI agents begin acting as economic participants, AuthPros develops the trust decisioning layer for evaluating whether an agent *should* transact, considering identity, delegated authority, and intent before financial execution. The responsible institution retains final authorization. - **Fundamental Rule**: Agents do not define trust. Institutions do. AuthPros designs explainable trust recommendations (`APPROVE`, `CHALLENGE`, `REVIEW`, `DENY`) with supporting decision evidence; technical cryptographic proof is a capability requiring verification at the deployed-runtime level. The institution (issuing bank, merchant, payment provider) preserves final financial authorization and execution. - **Headquarters**: São Paulo, Brazil (Operating Globally) - **Primary Contacts**: - General & Pilot Inquiries: contact@authpros.io - Strategic Partnerships: partners@authpros.io - Security & Vulnerability Reporting: security@authpros.io - **Official Social Profile**: https://x.com/authprosio --- ## Evidence and Product Status This public reference describes product direction, design contracts and illustrative examples. It is not proof of live issuer integration, contractual SLA, third-party certification, deployed MCP service, production-ready cryptographic proof, or independent security attestation. Verify actual availability and scope through AuthPros documentation and a technical review before relying on a capability. --- ## 2. Market Context & The New Decision Problem ### Market Growth - By 2030, a substantial portion of global consumer commerce is projected to be orchestrated or conducted by autonomous AI agents. - Gartner (2026): Observed up to 11% consumer willingness to let AI execute purchases autonomously in low-stakes categories. - Visa (2026): 23% of consumers express growing openness to GenAI handling payment transactions on their behalf. - Mastercard (2026): Projects that more than 1 in 10 online shoppers will routinely use AI agents to discover, negotiate, and pay by 2030. ### The Problem Traditional payment infrastructure (card schemes, instant payment rails, gateways) was engineered to determine whether a transaction *can* technically occur: - Is the account funded? - Are the credentials valid? - Is the rail available? Agentic commerce introduces an entirely new actor — the autonomous AI agent. An agent can technically initiate API calls and sign transactions, but payment rails cannot determine whether it *should* execute that specific purchase on behalf of that specific person or business: - Did the human principal intend this specific item, price, or vendor? - Is the agent operating within its delegated budget, expiration time, and category? - Has the agent hallucinated or been hijacked via prompt injection? - Does the transaction comply with the card issuer's or enterprise's internal governance rules? AuthPros provides the decision layer that answers these questions before execution. --- ## 3. Know Your Agent (KYA): The Five Evaluation Dimensions AuthPros unifies five core signals into a single trust decision: 1. **Identity (Who is acting?)**: - Resolves the identity of the AI agent, execution runtime, host environment, and cryptographic key material. - Decoupled from human identity to maintain clear separation of responsibilities. 2. **Delegated Authority & Mandates (What has been delegated?)**: - Evaluates the mandate binding the human principal to the agent. - Enforces scope, category bounds, maximum ticket values, cumulative limits, validity periods, and hierarchical sub-agent delegation. - Verifies operational mandate state: Created, Pending Consent, Active, Suspended, Revoked, Expired. 3. **Intent Verification (What was requested?)**: - Compares the human principal's instructions with the merchant's checkout proposal. - Assesses purpose, merchant authenticity, line-item justification, and divergence risk. 4. **Context & Intelligence (What does the environment tell us?)**: - Telemetry from agent behavior, velocity anomalies, IP/device posture, execution graphs, and historical patterns. 5. **Institutional Control & Policy (Does governance allow it?)**: - Applies issuer policies, merchant acceptance criteria, regulatory rules (e.g. anti-fraud, midnight limits on instant rails), and risk thresholds. --- ## 4. Architectural Invariants (Agentic Trust decision engine) The AuthPros Agentic Trust architecture specifies eight intended design invariants. Conformance in a particular deployed environment must be demonstrated independently: - **Invariant I0 — Canonical Separation**: Principal (human/organization), AgentIdentity (autonomous software instance), and Context are formally decoupled entities. - **Invariant I1 — Deterministic Decidability**: Pure, deterministic evaluation targeting p99 latency < 100ms. Same inputs and policy states yield reproducible recommendations. - **Invariant I2 — Strict Attenuation**: Sub-agent authority is mathematically bounded by parent authority ($Child \subseteq Parent$). An agent cannot delegate rights it does not possess. - **Invariant I3 — Non-Compensation**: A high commercial trust or affinity score can never compensate for or override an invalid, expired, or revoked mandate. - **Invariant I4 — Merchant Sovereignty**: Merchants retain the absolute right to decline or reject transactions independently of AuthPros recommendations. - **Invariant I5 — Tool vs Authorization**: Tool discovery mechanisms (such as MCP tool declarations) map available capabilities, but confer zero spend or execution authorization by themselves. - **Invariant I6 — Cryptographic Immutability**: The design calls for a tamper-evident, auditable `DecisionProof` artifact capturing provenance from signal to decision; completeness and cryptographic validation depend on the verified implementation. - **Invariant I7 — Passkey Invisible Step-Up**: The architecture contemplates asynchronous, out-of-band step-up authentication for `CHALLENGE` outcomes (e.g. FIDO2 / WebAuthn passkeys); production support and human-intent verification must be separately proven. --- ## 5. Payment Rails & Protocols — Architectural References AuthPros is designed as a rail-agnostic advisory layer above payment and commerce protocols. The following standards are architectural references or interoperability targets, not a verified list of live partner integrations or certification: ### Sovereign Instant Rails - **Pix (Banco Central do Brasil)**: EMV BR Code CRC16, Trava Noturna (nighttime limit policies), and CNPJ tax ID validation. ### Payment Scheme Rails - **Cards (EMVCo)**: 3DS 2.2.0, 3DS Requestor Initiated (3RI) Merchant-Initiated Transactions (MIT), Network Tokens. - **Mastercard Agent Pay**: Verifiable Intent Ed25519 signatures and Gateway v100 interoperability. - **Visa TAP**: Trust Assurance Protocol (AP-VISA-TAP-1 conceptual alignment). ### Agentic & Machine Protocols - **ACP (Agentic Commerce Protocol)**: Agent negotiation and checkout handshake. - **UCP (Universal Commerce Protocol)**: Cross-platform agent catalog and purchasing protocol. - **AP2 (Agent Payments Protocol)**: Mandate evidence exchange between consumer agents and merchant processors. - **x402**: HTTP 402 Payment Required machine payment standard with HMAC-SHA256 anti-replay tokens. - **MPP (Machine Payment Protocol)**: Continuous or micro-metered token-based payment streams. - **AP4M (Mastercard Agent Pay for Machines)**: High-frequency atomic velocity execution for machine transactions. - **Meta Muse / PAP (Personal Agent Gateway)**: Multi-turn goal-oriented commerce flows. - **Model Context Protocol (MCP)**: Tool exposure and context exchange for LLMs. - **A2A (Agent-to-Agent Cards)**: Autonomous multi-agent coordination and payment handoffs. --- ## 6. Knowledge Hub: Core Articles & Technical Concepts ### Guide 1: What is Agentic Commerce? Comércio agêntico (Agentic Commerce) represents the transition from human-driven point-and-click e-commerce to autonomous agent-driven commerce. AI agents discover products, negotiate conditions, evaluate inventory, and formulate purchase proposals. Key architectural distinctions: - **Discovery vs Decision vs Execution**: Discovering an item (agent tool) is distinct from deciding whether to buy it (trust decisioning), which in turn is distinct from moving funds (settlement rail). - **Why credentials are insufficient**: Providing an API key or card number to an agent creates systemic risk; agents require scoped mandates with temporal and financial caps. ### Guide 2: Trust in Agentic Commerce Trust in agentic commerce cannot rely on traditional human browser sessions, CAPTCHAs, or SMS OTPs. It requires automated, verifiable trust decisioning: - **Pre-transaction evaluation**: Signals must converge before the transaction hits the clearing house or financial rail. - **Auditable Provenance**: Every transaction must link back to the originating prompt, the specific agent execution thread, and the validated mandate. ### Guide 3: AI Agent Authorization & Delegated Mandates Authorization models for AI agents: - **Mandate structure**: Grantor ID, Agent ID, Validity Window, Currency, Total Limit, Per-Transaction Limit, Merchant Whitelist / MCC Category Codes. - **Lifecycle management**: Dynamic capability to suspend, adjust limits, or immediately revoke mandates via API or human dashboard. ### Guide 4: Intent Verification in Agentic Transactions Verifying intent ensures that the agent's actions correspond to what the principal requested, protecting against prompt injection, adversarial drift, and algorithmic errors: - **Semantic evaluation**: Matching prompt embeddings and constraints with merchant cart manifests. - **Discrepancy thresholds**: Automatically triggering `CHALLENGE` or `REVIEW` when discrepancies in merchant identity, price, or item specifications are detected. --- ## 7. Model Context Protocol (MCP) Integration AuthPros Agentic Trust will expose a remote Model Context Protocol endpoint (`https://mcp.authpros.io/agentic-trust`, Streamable HTTP, JSON-RPC; status: Coming soon · pilot access, not live yet) allowing agents to interface with the decision engine: ### Configuration ```json { "mcpServers": { "authpros-agentic-trust": { "url": "https://mcp.authpros.io/agentic-trust", "headers": { "Authorization": "Bearer ${AUTHPROS_API_KEY}" } } } } ``` ### Available Tools 1. `evaluate_transaction`: - Evaluates a purchase proposal against mandate limits, intent match, and institutional policies. - Parameters: `transaction_id`, `mandate_id`, `amount`, `currency`, `merchant`, `rail`. - Returns: Deterministic decision (`APPROVE`, `CHALLENGE`, `REVIEW`, `DENY`), risk score, reason codes, and tamper-evident `DecisionProof`. 2. `get_mandate`: - Inspects mandate parameters, status, and remaining spend balance (`authorized - consumed - reserved`). - Parameters: `mandate_id`. 3. `check_policy`: - Validates whether a hypothetical transaction satisfies institutional rules without consuming balance or committing state. 4. `verify_evidence`: - Verifies cryptographic signature and immutability of an issued `DecisionProof`. 5. `register_agent`: - Enrolls an agent public key and runtime fingerprint into the tenant identity directory. --- ## 8. Agentic Trust API v3 Specification (Advisory Quickstart) ### Evaluation Request (`POST /v3/evaluate`) ```json { "api_version": "3", "transaction_id": "tx_7f8a9b2c3d4e", "operation": "EVALUATE", "resource": "mandates/man_corp_travel_2026", "authority_id": "man_corp_travel_2026", "idempotency_key": "idemp_5c91e0a8b72d", "agent_identity": { "agent_id": "agent_procurement_42", "initiator_agent_id": "Human_Principal_Exec_01" }, "dat": "dat_", "payment": { "type": "CARD", "payload": { "amount": 85.00, "currency": "USD", "merchant": "Cloud Compute Corp", "merchant_tax_id": "US-EIN-12-3456789", "card": { "scheme": "VISA", "tokenType": "NETWORK_TOKEN", "panMasked": "4111********1111" } } }, "evidence": [ { "scope": "EVALUATIVE", "type": "EMV_3DS_NETWORK_TOKEN", "assurance_level": "PASSKEY_FIDO2", "verified": true } ] } ``` ### Advisory Decision Response ```json { "decision_id": "dec_94a8c3d2e1f0", "recommendation": "APPROVE", "confidence_score": 0.985, "latency_ms": 38, "evaluations": { "identity": "VERIFIED", "authority": "VALID", "intent": "ALIGNED", "transaction": "IN_SCOPE", "institutional_policy": "PASS" }, "evidence_id": "evi_1234abcd5678", "tamper_proof_signature": "ed25519:3a9f...b48c" } ``` --- ## 9. Frequently Asked Questions (FAQ) **Q: What does AuthPros decide?** A: AuthPros is developing advisory evaluation of agent identity, delegated authority, transaction intent, contextual signals, and institutional policy before execution. Outputs and cryptographic evidence guarantees vary by verified deployment scope. **Q: Does AuthPros authorize or execute payments?** A: No. AuthPros provides a trust decision recommendation. The financial institution (issuing bank, acquirer, or merchant) retains legal authorization and settlement execution control. **Q: How is delegated authority evaluated?** A: The decision model verifies the mandate, permitted scopes, limits, and expiration against the requested merchant, amount, purpose, and hierarchical sub-agent permissions. **Q: Does referencing protocols (e.g. Visa TAP, Mastercard Agent Pay, x402) imply partnership?** A: No. Protocol and vendor references indicate ecosystem standards and architectural interoperability targets; they do not imply direct commercial endorsement or partnership. **Q: How can enterprises pilot AuthPros?** A: Organizations can contact contact@authpros.io to schedule technical alignment and pilot access. --- ## 10. Security, Privacy, and Compliance - **Data Minimization by Design**: The intended architecture minimizes exposure to sensitive PAN and raw payment credentials. A zero-PAN storage assertion requires scoped data-flow and production evidence. - **Encryption**: TLS and encryption at rest are security requirements; exact protocol/cipher choices and coverage are not represented as universally verified without environment-specific evidence. - **Audit Evidence**: Immutable provenance records are an architectural objective; real coverage, retention, integrity and audit suitability require deployment and third-party evidence. - **Responsible AI**: Design objective of explainable and deterministic decision recommendations without unauthorized overrides of legal mandates. - **Compliance Process**: Security documentation can be discussed with qualified enterprise customers under appropriate confidentiality procedures. No SOC, ISO, PCI-DSS certification or attestation is claimed without independently verifiable evidence.