AI PAYMENT SECURITY LIBRARY
Answers for every team between an AI agent and a payment.
Use-case playbooks, threat explanations, protocol guides and machine-readable integration resources.
FREE INTERACTIVE TOOLS
Get an answer, artifact or action plan
Useful without an account. Built to move a real architecture decision forward, not merely to collect an email address.
Normalize native payment syntax
Convert three agent-payment formats into one deterministic, explicitly non-executable authority envelope.
Open protocol bridge →CRYPTOGRAPHIC AUTHORITY LABProve signed purchase authority
Generate an ephemeral key, sign a purchase and independently verify the portable decision receipt.
Run the proof →3-MINUTE ASSESSMENTScore the payment boundary
Answer ten control questions and copy a prioritized security report.
Assess risk →MANDATE BUILDERGenerate testable purchase authority
Create fresh AP2, TAP, UCP, x402, ACP or custom JSON and run it live.
Build a mandate →LIVE ATTACK LABAttack a purchase five ways
Test overspend, seller substitution, expiry, consent and prompt injection.
Run live checks →EMBEDDABLE DISCLOSUREShow the payment boundary
Add an accessible disclosure badge to a protected checkout or integration guide.
Get badge code →VERSIONED PRODUCT EVIDENCE
Verify the claims, boundaries and protocol coverage
Human-readable matrices link each material claim to public contracts and versioned JSON for machines.
Verify the shipped SDK bytes
Inspect versioned SHA-256 checksums and the first-party SPDX software bill of materials.
Open release manifest →BOUNDED FORMAL MODELExecute the state-machine invariants
Review the TLA+ specification and bounded explorer results for budget, consume-once, UNKNOWN, key rotation and tenant isolation.
Inspect formal evidence →DEPLOYMENT ASSURANCEVerify the customer-controlled Gate
Run the offline Doctor against strict evidence and inspect conservative Node, serverless and Kubernetes templates.
Open Gate assurance →CONTROL EVIDENCEPayment-authority control matrix
Inspect deterministic policy, cryptographic authority, replay, reservation and processor-boundary evidence.
Inspect controls →PROTOCOL EVIDENCEExact support and responsibility matrix
Separate native verification, normalized projections and responsibilities that remain external.
Inspect protocols →PUBLIC TEST PROFILERun the conformance vectors
Validate compatible policy decisions against public schemas and positive and negative test cases.
Open conformance suite →IMPLEMENTATION GUIDES
Move from architecture to an enforceable boundary
Connect authority to one provider attempt
Use a durable customer-side gateway, exact bindings and conservative reconciliation.
Integrate Stripe →MCP PAYMENT AUTHORIZATIONKeep payment power outside the model
Connect MCP tools and paid APIs to a consume-once trusted gateway.
Read guide →SECURITY CHECKLISTTwelve controls before money moves
Review authority, checkout binding, concurrency, replay and unknown provider outcomes.
Use checklist →MERCHANT GUIDEAI-agent payment security
Understand the seven checks that belong between autonomous selection and payment submission.
Read guide →AP2 VALIDATIONValidate runtime mandate evidence
Bind amount, payee, checkout, freshness and replay protection.
Read checklist →STABLE DEFINITIONSAgentic payments glossary
Give teams and AI systems one precise vocabulary for identity, authority, execution and settlement.
Open glossary →DECISION COMPARISONS
Choose the right layer and protocol
AI-agent payment authority vs. direct-debit mandates
MandateShield protects delegated AI purchases. It is not SEPA mandate collection, direct-debit processing, electronic signature or bank-payment software.
Compare →CONTROL COMPARISONAI payment authorization vs. fraud detection
Fraud tools ask whether a payment looks suspicious. Agent-authorization controls ask whether the person permitted this exact autonomous purchase.
Compare →PROTOCOL COMPARISONAP2 vs. TAP vs. x402 for agent payments
AP2, Visa Trusted Agent Protocol and x402 solve different layers of agentic commerce. A payment boundary can normalize the facts needed for execution-time authority.
Compare →ARCHITECTURE DECISIONBuild vs. buy AI payment guardrails
A homegrown rule can compare an amount. A production execution boundary must also normalize protocols, fail closed, stop replay and preserve explainable evidence.
Compare →PROTOCOL COMPARISONMachine Payments Protocol (MPP) vs. x402
MPP and x402 both use HTTP 402 for machine payments, but they differ in payment-method scope and integration semantics. Neither replaces delegated purchase authority.
Compare →CONTROL COMPARISONAI payment authorization vs. virtual-card controls
Virtual cards constrain a payment credential. AI payment authorization proves whether the agent's exact final purchase remains inside delegated intent.
Compare →TRUST COMPARISONAI-agent identity vs. payment authority
Agent identity answers who is acting. Payment authority answers whether that actor may execute this exact purchase now.
Compare →RUNTIME SECURITY COMPARISONPayment policy check vs. execution integrity
A policy ALLOW says a proposed purchase fits the rules. Execution integrity binds that decision to one exact downstream request, a single-winner MandateShield claim and a checked terminal outcome.
Compare →HONEST ARCHITECTURE COMPARISONProvider-native controls vs. MandateShield outcome verification
AWS, Crossmint, Mastercard, Stripe and AP2 already provide important agent-payment controls. MandateShield adds value only when its narrow caller-independent reconciliation and evidence boundary fills a remaining gap.
Compare →STACK COMPARISONAI payment authorization vs. payment processing
Authorization policy decides whether an agent may attempt a purchase. Payment processing authorizes, captures and settles money on a payment rail.
Compare →SOLUTIONS
Execution controls by operating model
Outcome verification for AI agents using Stripe or x402
Bind a supported provider request, grant one fresh online permit claim, and verify the terminal outcome without trusting the calling agent's report.
Read solution →Ecommerce merchantsAccept AI-agent orders without trusting the model
Reserve authority for autonomous checkout totals, merchant identity and user-approved limits, then require gateway consumption before creating a charge.
Read solution →MarketplacesPolicy enforcement for multi-merchant agent checkout
Control seller substitution, price drift and replay across marketplace purchases initiated by AI agents.
Read solution →Payment providersPre-authorization control for agentic payments
Add user-mandate enforcement before payment processing, alongside existing fraud and risk systems.
Read solution →Fintech and complianceExplainable controls for autonomous transaction authority
Produce deterministic AI-payment decisions, reason codes and minimal receipts for operational oversight.
Read solution →Paid APIs and MCP toolsPayment controls for paid APIs, MCP tools and machine services
Authorize provider, resource, exact price and cumulative budget before an AI agent pays for an API call or MCP tool.
Read solution →Enterprise procurement agentsAI procurement agent spending controls
Give procurement agents enforceable supplier scopes, exact transaction limits and concurrent daily or monthly budgets.
Read solution →THREAT LIBRARY
Failures to stop before money moves
Stop AI-agent overspend before payment
Enforce a hard maximum against the final amount, including checkout changes, fees and currency boundaries.
Read threat guide →Merchant substitutionPrevent merchant substitution in agentic checkout
Bind an autonomous purchase to the intended seller or approved merchant scope before execution.
Read threat guide →Prompt-injection payment containmentContain prompt injection at the payment boundary
Assume instruction overrides can succeed, use heuristic signals only for review, and contain payment impact with deterministic policy outside the model.
Read threat guide →Payment replay attacksEnforce consume-once authority before payment submission
Consume each MandateShield decision once, reject a second permit redemption, and preserve provider idempotency as a separate requirement.
Read threat guide →Authorization-to-execution gapMediate the AI-payment authorization-to-execution gap
Bind an approved autonomous purchase to one supported provider request, one fresh online permit claim and an outcome checked by MandateShield without trusting the caller report.
Read threat guide →Consent driftKeep AI purchases inside current user consent
Reject missing, stale or weakly bound authority when checkout facts change after the user delegates a purchase.
Read threat guide →Concurrent agent budget exhaustionStop concurrent AI agents from overspending one shared budget
Use atomic reservations so parallel agent purchases cannot all pass the same stale daily, monthly or lifetime budget check.
Read threat guide →FOR DEVELOPERS AND AI SYSTEMS