Pay with confidence: How Solv Labs built verifiable, auditable agent payments on Amazon Bedrock AgentCore payments
Amazon Web Services Patrick Duffy ● Covered by 3 sources
Solv Labs built agent payments on AWS that can be checked after the fact. Each payment carries proof, a risk price, and an audit trail before money moves.
Based on reporting by Amazon Web Services, Patrick Duffy — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Solv Labs says it has built an AI agent payment workflow on Amazon Bedrock AgentCore payments that does something enterprises have wanted for a while: make every machine-driven payment provable. The system sits on top of AgentCore payments and adds two layers of governance, ORACLE, Solv’s policy engine, and ICME PreFlight for compliance checks.
The core idea is simple enough. Before any money leaves, ORACLE decides whether the action is allowed or needs review. ICME’s verification layer turns the policy check into a privacy-preserving proof that someone else can verify without seeing the policy or the transaction details. Then an integrity service inside an AWS Nitro Enclave signs the execution record, and a risk engine assigns each transaction its own risk multiplier.
Only after that does settlement happen, with on-chain routing through Coinbase. Solv says the full path, from pre-authorization through governance to settlement, takes under four seconds per transaction. That timing matters because the whole pitch here is that control doesn’t have to slow agents down. It can ride along with them.
The company frames this as a response to a basic enterprise problem: when an autonomous system moves money, the operator needs more than a generic trust statement about the organization behind it. It needs a record tied to the specific payment, showing what policy it cleared, what constraints it met, what risk it carried, and what enclave signed it. Model cards and SOC 2 reports don’t do that job. A transaction-level record does.
Solv says each governed payment produces a signed evidence record that can be checked independently, with the policy result, the proof, the enclave attestation, the risk price, and the settlement artifacts all bound together. AgentCore payments also enforces per-session spending limits and keeps the whole workflow inside the same environment as the agent, so there isn’t a second control plane sitting off to the side causing the usual headaches.
The broader bet is that governed agent payments stop being a research exercise once the plumbing exists: AgentCore payments, AWS Automated Reasoning Checks, Nitro Enclaves, and x402. Solv’s version of the pitch is not that agents are magically safe. It’s that every payment can at least be accounted for, which is a lot more useful when real money is involved.
My take — AI-written commentary, not fact-checked reporting
This is the right direction, because “the agent said so” is not an audit trail. The industry has spent years pretending logs are governance; this at least treats proof as part of the product. If AI is going to handle money, the bar should be boring, strict, and a little paranoid.
Read more about this at: Amazon Web Services