Verifiable Agent Payments Built on Amazon Bedrock for Regulated Environments

Solv Labs, in collaboration with ICME Labs, has built an advanced AI agent-payments workflow leveraging Amazon Bedrock AgentCore payments. This system ensures that every payment made by autonomous agents is governed at execution time, with each transaction producing a verifiable, auditable record. The workflow integrates multiple layers of security and verification, including an ORACLE policy engine, hardware attestation via AWS Nitro Enclaves, and risk-based pricing.

The architecture involves a sequence of components that process each payment: the ORACLE evaluates the proposed transaction against policies, PreFlight provides privacy-preserving policy verification proofs, Nitro Enclaves cryptographically attest to the integrity of the execution, and a risk engine assigns a risk multiplier to each transaction. Only after passing through these layers does the payment settle on-chain via Coinbase, with all steps completed within four seconds.

This approach addresses a critical enterprise challenge: providing a durable, transaction-level record that links each payment to its authorization policy, constraints, and risk assessment. Such records are essential for audits, dispute resolution, and regulatory compliance, especially in environments where autonomous systems handle financial transactions.

The system's design ensures that all evidence—policy checks, hardware attestations, risk pricing, and settlement details—is cryptographically signed and anchored on a public blockchain, enabling third-party verification without exposing sensitive policy or transaction data. This infrastructure supports scalable, real-time governance, reducing review efforts to exception cases and fitting within existing operational budgets.

Overall, this implementation demonstrates how recent advances in AI infrastructure, hardware security, and cryptographic verification can be combined to enable trustworthy, compliant autonomous agent payments at enterprise scale.