AI Agent Purchase Liability Unsettled as Agentic Commerce Surges
Agentic commerce is growing rapidly without a settled liability framework, meaning merchants, users, and insurers are absorbing risks that card network rules do not yet address. Insurance and technical audit approaches are emerging to fill that gap, but none are yet standardized.
Agentic shopping traffic recorded a 100x surge over 45 days following releases of consumer AI agents, per cybersecurity firm Cheq, and this holiday season is expected to be the first real test of which brands can effectively market to AI agents. No existing Visa or Mastercard rule assigns chargeback liability to an AI platform when an agent makes an unintended purchase, leaving merchants bearing that liability. Card networks have introduced technical standards to carry an intent mandate from agent to merchant but have not published specific liability rules. Merchants are responding differently: Amazon and eBay have blocked AI shopping agents to maintain control over their ad businesses and checkout flows; Target holds users 100% liable for whatever their agents do; Expedia declared on 24 September that it stays merchant of record for hotel bookings made via its Muse agent, keeping dispute ownership with the merchant. Morning Consult data shows consumers are as likely to be swayed by an AI agent's recommendation as by an influencer, but remain uncomfortable letting AI make purchases autonomously. On the insurance side, Harbor emerged from stealth with a $10M seed round led by Base10Partners, Abstract VC, and 8vc, covering six named perils: hallucinations, action errors, bias and discrimination, contract and SLA failure, data leakage, and IP infringement and defamation. A Carnegie Mellon paper found that existing agent logs, tested on OpenClaw, left zero of four tested policy violations provable because approvals and permissions were buried in unstructured chat text; the paper proposes a blockchain-backed verifiable agent insurance model that achieved 42.2% proven violations in tests, with an estimated 83% achievable using structured logs at under one cent in blockchain fees per session.
A Carnegie Mellon paper found zero of four policy violations provable from OpenClaw's logs because approvals were buried in unstructured text, and proposed a blockchain-backed model achieving 42.2% proven violations.
Cheq recorded a 100x surge in agentic shopping traffic over 45 days; eBay joined Amazon in blocking AI agents to protect ad and checkout businesses; and Odessia launched as an accredited travel agent AI, showing how commercial registration shifts liability ownership.
JP Morgan Chase blocked external AI agents from directly accessing customer accounts as of October 15, a unilateral institutional response to an environment where no Visa or Mastercard rule yet assigns chargeback liability to an AI platform when an agent makes an unintended purchase.
Card networks have built technical standards for passing an intent mandate from agent to merchant but have not published specific liability rules or created an AI-agent chargeback carveout.
Harbor, a startup founded by four Stanford graduates, raised a $10M seed round led by Base10Partners, Abstract VC, and 8vc on October 13 to offer insurance covering AI agent failures, with named-peril policies for hallucinations, action errors, bias and discrimination, contract and SLA failure, data leakage, and IP infringement and defamation.
The funding converts what Microsoft CEO Satya Nadella had called a prediction, that insurance would price the liability risk, into a live commercial product. Card networks have still not published specific liability rules for agent-initiated purchases, and merchants remain split, with Target holding users fully liable, Expedia staying merchant of record inside its Muse agent, and Amazon blocking agent access entirely.
Debate reported on who bears liability for chargebacks when AI agents autonomously make purchases; Amazon, Target, and Expedia positions documented
A coalition of banks published a report on October 8 confirming that no Visa or Mastercard rule assigns chargeback responsibility to an AI platform when an agent makes a mistaken purchase, leaving merchants as the default liable party.
The Merchant Advisory Group added that agents might absorb that liability as a premium commercial feature, though it noted the cost could become prohibitive. Traxtech reported 70% of consumers are comfortable with AI agent purchases, with payment security their top concern at 32%.
Traxtech reported 73% of shoppers use AI in purchasing journeys; payment security is top consumer concern at 32%
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free