Solv Labs' AgentCore payments give AI agents auditable transactions in 4 seconds

Solv Labs, in partnership with ICME Labs, has unveiled a production-ready agent-payments workflow built on Amazon Bedrock AgentCore payments, a capability launched in May 2026 with Coinbase and Stripe. The system directly tackles the enterprise governance gap for autonomous agents that move real money. Every transaction passes through three governance components: ORACLE, Solv's policy engine that pre-authorizes spending; an integrity service running inside an AWS Nitro Enclave that attests to the execution; and a risk engine that prices each transaction individually. The full cycle—from pre-authorization to on-chain settlement via Coinbase—completes in under four seconds, well within latency budgets for agentic workloads. Each payment generates a durable audit trail that ties the specific action to the policy that authorized it, the constraints satisfied, and the risk carried, making disputes and audits straightforward.

The architecture is notable for achieving verifiability without relying on the operator's word. Policy checks from ICME extend AWS Automated Reasoning Checks to be privacy-preserving and independently verifiable, while AWS Nitro Enclaves act as a per-transaction hardware attester. Components cross trust boundaries only through signed, hash-bound artifacts, ensuring the evidence flow holds even if the service is compromised. This pattern was impractical until the convergence of AgentCore payments, Automated Reasoning Checks, Nitro Enclaves, and the x402 payment standard. For enterprises in regulated environments, it transforms agent payments from a research experiment into an implementation choice, offering transaction-level proof that model cards and SOC 2 reports cannot provide. Agents can now spend on web content, APIs, MCP servers, and even other agents, with the same controls operators already use—but with a formally verifiable record for every dollar moved.