AI/Nov 13 edition/Reporting & analysis
AgentsBusinessInfrastructure

AgentsAutonomy & tool use

Jamie Dimon Bars Outside AI Agents From Chase Accounts as Stripe and Card Networks Build Agent Payment Rails

JPMorgan is limiting outside AI agents, and some agent purchases made with virtual cards are being declined. Meanwhile, Stripe, Mastercard and Visa are building payment credentials designed for agents. How an agent's purchase is authenticated is becoming the key dividing line.

From the WSJ · Technology article of November 13, 2026 and cited sources: JPMorgan is restricting outside agents. [4] [9] [3] [10]

SOURCES[4] Fortune: Jamie Dimon on AI agents fortune.com[9] Crowdfund Insider: JPMorgan restricts external AI agents crowdfundinsider.com[3] CryptoBriefing: JPMorgan blocks some AI agent purchases cryptobriefing.com

AI-assisted, not human-edited. Methodology Corrections Who owns this site

KEY FACTS
What happened
JPMorgan is restricting outside agents. [4] [9] [3] [10]
Original source
WSJ · Technology article, independent perspective
Brief by
AT Pulse research desk AI-assisted; publisher and contact on the desk page
When
November 13, 2026 article published
Evidence
11 sources across 10 sites

Reach: not measured yet

THE CORE IDEAS3 TAKEAWAYS
01

JPMorgan is restricting outside agents. On the bank's Q3 call, Jamie Dimon said external AI agents won't get into Chase customer accounts because the risk is too high. Crypto-news outlets also report that Chase declines some agent purchases because it treats the agents' one-time virtual cards as commercial cards. [4] [9] [3] [10]

02

Payment companies are building credentials specifically for agents. Stripe and OpenAI's open Agentic Commerce Protocol uses a Shared Payment Token limited to one merchant and one cart total, and the merchant remains the merchant of record. Mastercard's Agentic Tokens tie a credential to a specific agent and add spending limits and consent steps. Visa's Trusted Agent Protocol helps merchants tell trusted agents apart from bots. [6] [7] [5] [12]

03

Consumer demand and accountability are still unresolved. In March 2026, OpenAI scaled back in-chat Instant Checkout after low conversion and heavy operational load. A JPMorgan Payments executive noted that card liability rules assume only three parties, so it is unclear who absorbs the loss when an agent orders the wrong item. [8] [11]

WHY IT MATTERS

major payment firms are building agent-specific tokens while the largest US bank limits outside agents. Implication: merchants that rely on virtual cards may lose sales, and agent builders face bank-by-bank rules until liability standards settle.

Sources are verified; writing the dossier now.

Executive brief

The largest US bank is blocking AI agents while payment companies race to support them. JPMorgan Chase reportedly treats the virtual cards that AI agents use as commercial cards, so some agent purchases get declined at checkout (CryptoBriefing). Jamie Dimon has also said outside agents won't get into Chase customer accounts (Fortune). Stripe, meanwhile, wants to be the core infrastructure for agentic commerce (WSJ). Only the WSJ feed excerpt was reviewed, so its details are not reflected here.

What changed and event timeline

  1. Mastercard launches Agent Pay

    Mastercard introduced "Agentic Tokens" built on its existing tokenization system, with programmable spending controls ().

  2. Stripe and OpenAI release ACP

    Stripe ran Instant Checkout in ChatGPT and open-sourced the Agentic Commerce Protocol, which included a new Shared Payment Token ().

  3. Stripe ships the Agentic Commerce Suite

    This low-code product lets merchants sell through several AI agents with one integration. Named adopters include Coach, URBN, Wix and Etsy ().

  4. OpenAI scales back Instant Checkout

    Purchases moved into third-party apps. Low conversion and operational burden were cited as reasons ().

  5. Mid Oct 2026

    Dimon warns about "rogue agents."

    On JPMorgan's Q3 call, he ruled out letting external agents into customer accounts, saying it was too risky (;).

  6. WSJ describes the split

    WSJ frames the story as Stripe backing agentic commerce while JPMorgan stays skeptical ().

Capabilities and access

  • Stripe ACP / Shared Payment Token (SPT): An open standard.
  • Agentic Commerce Suite: Stripe manages the SPTs across multiple agents for merchants (Stripe).
  • Mastercard Agentic Tokens: Credentials tied to a specific agent, with spending limits and consent flows (Eco).
Read the full section
  • Stripe ACP / Shared Payment Token (SPT): An open standard. An SPT lets an agent start a payment without exposing the card, and is limited to one merchant and one cart total (Stripe).
  • Agentic Commerce Suite: Stripe manages the SPTs across multiple agents for merchants (Stripe).
  • Mastercard Agentic Tokens: Credentials tied to a specific agent, with spending limits and consent flows (Eco).
  • JPMorgan: Read-only data access through approved aggregators such as Plaid. External agents can't act on retail accounts (KuCoin/CryptoBriefing).

Technical analysis for researchers and developers

  • Credential design is the main fault line. Agents using one-time virtual card numbers can be flagged as commercial cards and declined.
  • Merchant control: In ACP, the merchant stays the merchant of record. It accepts or declines orders, handles tax and processes returns (Stripe).
  • For developers: Build on tokenized flows from Stripe or the card networks, not on screen-scraping or logging in with a user's credentials.
Read the full section
  • Credential design is the main fault line. Agents using one-time virtual card numbers can be flagged as commercial cards and declined. Mastercard CEO Michael Miebach reportedly said such cards "fundamentally break" a transaction and favors tokens built for agents (CryptoBriefing).
  • Merchant control: In ACP, the merchant stays the merchant of record. It accepts or declines orders, handles tax and processes returns (Stripe).
  • For developers: Build on tokenized flows from Stripe or the card networks, not on screen-scraping or logging in with a user's credentials.
  • Evaluation: No public, reproducible data on fraud or authorization rates for agent payments turned up in the reviewed sources.

Claims and evidence

  • JPMorgan classifies agent virtual cards as commercial cards: Reported by CryptoBriefing; no bank statement was found (CryptoBriefing).
  • Schwab, Citi and Wells Fargo also restricted agents (Oct 13–20): Comes from the same secondary chain, with no independent confirmation (KuCoin).
  • Dimon's remarks: Covered by several outlets (Fortune; Crowdfund Insider).
Read the full section
  • JPMorgan classifies agent virtual cards as commercial cards: Reported by CryptoBriefing; no bank statement was found (CryptoBriefing).
  • Schwab, Citi and Wells Fargo also restricted agents (Oct 13–20): Comes from the same secondary chain, with no independent confirmation (KuCoin).
  • Dimon's remarks: Covered by several outlets (Fortune; Crowdfund Insider).
  • Low uptake of Instant Checkout: Based on The Information's reporting and Shopify's comments (Retail Gazette).
  • Stripe's capabilities and adopters: Reported by Stripe itself (Stripe).
  • A figure saying agent chargebacks run about 40% higher: Appeared in a search summary but couldn't be traced to a source, so it is excluded.

Context and prior work

  • In May 2026, a JPMorgan Payments executive said the bank had stayed quiet on purpose.
  • Visa announced a Trusted Agent Protocol in October 2025 to help merchants tell trusted agents from bots (Trending Topics).
  • Meta's Muse assistant launched in September 2026 and connected to bank accounts through Plaid (Fortune).
Read the full section
  • In May 2026, a JPMorgan Payments executive said the bank had stayed quiet on purpose. He called today's tools "AI-embedded commerce" rather than truly autonomous agents (Stellagent).
  • Visa announced a Trusted Agent Protocol in October 2025 to help merchants tell trusted agents from bots (Trending Topics).
  • Meta's Muse assistant launched in September 2026 and connected to bank accounts through Plaid (Fortune).

Limitations, safety and contested findings

  • Liability: Card liability rules assume three parties: issuer, merchant and consumer.
  • Security: Fortune cites a Cloud Security Alliance report of criminals using AI agents to breach more than 27 organizations (Fortune).
  • Demand: OpenAI's pullback shows consumer demand is still unproven (Retail Gazette).
Read the full section
  • Liability: Card liability rules assume three parties: issuer, merchant and consumer. When an agent orders the wrong item or misreads an order (for example, buying five phones instead of one), no legal party clearly absorbs the loss (Stellagent).
  • Security: Fortune cites a Cloud Security Alliance report of criminals using AI agents to breach more than 27 organizations (Fortune).
  • Demand: OpenAI's pullback shows consumer demand is still unproven (Retail Gazette).
  • Contrast: Germany's Scalable Capital lets ChatGPT and Claude help trade, with user approval (Fortune).

Business and practitioner implications

  • Merchants: Agent purchases paid with virtual cards may be declined by some issuers, which means lost sales.
  • Use approved aggregators instead of logging in with users' credentials.
  • Banks and networks: Whoever writes the liability rules for agent payments will shape the market.
Read the full section
  • Merchants: Agent purchases paid with virtual cards may be declined by some issuers, which means lost sales. Network tokens or SPTs are the safer path (KuCoin).
  • Agent builders: Expect bank-by-bank policies. Use approved aggregators instead of logging in with users' credentials.
  • Banks and networks: Whoever writes the liability rules for agent payments will shape the market. JPMorgan expects merchant-run agents on merchants' own sites to come first (Stellagent).
FOLLOW THE EVIDENCE

The source trail.

Sources (11)
A LITTLE LESS NOISE. A LOT MORE CONTEXT.

Stay curious.
Follow the evidence.

Independent perspectives, the original sources, and room for the questions that don't have easy answers.

How we build the brief
Connect with us

Find us where you already read.