Agent Bond: escrowed performance bonds for B2B agent vendors

No one is yet selling vendor-posted, log-adjudicated performance bonds for B2B finance agents. The mechanism exists only in research papers (the verifiable agent insurance paper and the Agentic Risk Standard) and in crypto hackathon projects aimed at simple API calls. The commercial demand is being captured by insurance startups instead (Klaimee, Mount, and AIUC's certification), and insurance fits existing procurement and is easier on vendor capital than a cash bond. The best opening is to be the neutral verification and settlement layer for accounts payable and receivable agents. That means a domain-specific promise spec, plus a signed violation report that a buyer, an insurer or a bank escrow can act on without arguing. It would work as a first-loss bond under insurance, or as the route for small vendors that can't get insured. Bank escrow rather than stablecoins only, and legally enforceable contract templates, would set it apart from the crypto projects. The risks are that the spending-policy vendors (Fireblocks, Sardis, Newton) could add signed attestations cheaply, and that insurers could make bonds unnecessary.

• Klaimee (YC P26)

  Raised $5.5M in July 2026. Sells insurance-backed performance warranties, financial guarantees and liability policies for AI agents after behavioral probe testing. It solves the same procurement-indemnity problem through insurance instead of an escrowed bond, which makes it the closest commercial substitute.

• Mount (YC)

  Building an AI-agent insurance carrier. Its first product is an AI liability policy for mid-market and enterprise companies deploying autonomous agents, so it competes for the same buyer reassurance budget.

• AIUC (Artificial Intelligence Underwriting Company)

  Raised a $40M Series A in September 2026. Its AIUC-1 is a SOC 2-style audit and certification standard for agents that feeds insurance underwriting. It could set the de facto standard for agent promises that a promise spec would have to fit.

• Verifiable Agent Insurance (arXiv paper)

  Research framework that is nearly identical to Agent Bond. Providers register promises about agent behavior and hold a reserve in a smart-contract escrow, and verified user challenges trigger payouts from it. Not a deployed product.

• Agentic Risk Standard (Microsoft Research, Columbia, DeepMind, Virtuals, t54 Labs)

  Open-source settlement standard proposal with escrow vaults, required provider collateral and optional underwriting for agent tasks that move money. It is research and a standard, not a commercial product, but it could become the reference design.

• ArcSLA

  Hackathon project where providers stake USDC against published SLA terms. A missed response window automatically slashes the stake and refunds the caller. It shows the mechanism for simple API calls, not finance workflows.

• Bazaar

  Hackathon project where AI service sellers post a USDC bond. Under-delivery judged by validator consensus is auto-slashed and the buyer refunded on-chain.

• Fireblocks x402 agent spending controls

  Enforces agent payment limits as policy rules outside the agent. It overlaps with the promise-enforcement side but does not issue violation reports or bonds.

• Sardis

  Natural-language spending policies with a real-time policy firewall and mandate verification with evidence. A possible starting point for a promise spec and monitor.

• Newton

  Enforces caps, payee allowlists and mandate scope per transaction using Rego policies, with an auditable record. Close to the promise policy language idea.

• agent-escrow (open source)

  Open-source milestone escrow with programmatic verification and conditional payment release for multi-agent work. Escrow tied to checks rather than vendor bonds.

Checked 2026-10-14