The biggest banks in the United States are drawing hard boundaries around AI agents — and the line runs straight through the customer's account. On JPMorgan's third-quarter earnings call on October 13, 2026, CEO Jamie Dimon said the bank will not let outside AI agents operate inside customer accounts on a client's behalf. Days earlier, Citi CEO Jane Fraser struck a similar note at the Sibos banking conference in Miami, demanding strict identity and authentication controls before any agent is allowed to execute. For startups building the agentic commerce layer, this is the distribution gate that matters most.
What Dimon actually said — and what the ban covers
Dimon told analysts: "We're not going to allow just any outside agent to come in inside the company, because it's just too risky for clients and for JPMorgan." He warned of "rogue agents," adding that "there could be thousands of them," and framed the issue as one of liability: customers who hand authority to outside agents should understand "there's a liability transfer."
The restriction has a precise shape:
- Blocked: Outside agents cannot act on a customer's behalf or log in with a customer's credentials. They cannot send money or pay bills.
- Allowed: Customers can still share account data — such as balances — with outside agents through authorized open-banking connectors like Plaid, on a read-only basis.
- Named: The tools in question include Meta's Muse and Instinct, an AI agent startup from Spear Street Technology that reportedly reached a $10 billion valuation last month.
A JPMorgan spokeswoman confirmed to CNBC that the bank gives no timeline for opening direct account operation to outside agents. Dimon's blunt framing: a rogue agent making the same bad call across thousands of accounts at once is a different order of risk than a single employee's mistake.
Citi's different answer: governance, not prohibition
Fraser, speaking at Sibos 2026 in Miami on September 30, did not announce a blanket ban. Instead, she argued that autonomous AI agents need rigorous identity verification and authentication before they can execute transactions — and she put Citi's internal house in order first.
The centerpiece is ARC, Citi's internal AI governance platform, introduced in 2026 and mandatory for all internal AI agents. "You cannot create an agent without going through ARC," Fraser said. No employee can spin up an AI agent without routing it through ARC's control and observability layer, keeping every agent's actions traceable to a human owner.
Fraser's emphasis suggests the issue isn't agents per se but unverified ones — a meaningfully different posture from JPMorgan's lockout, and a possible template for how banks eventually admit outside agents: verified identity first, execution second.
The industry-wide pattern
The big-bank earnings calls this month formed a chorus on AI risk:
- Bank of America CEO Brian Moynihan said the bank rushed to shore up vulnerabilities in its codebases after the release of powerful new models, hiring 1,000 external consultants over the last quarter to patch issues.
- Morgan Stanley CEO Ted Pick said the firm is investing "whatever it takes" to protect client assets, because "you can't engage head-to-head with AI out of your garage."
- Goldman Sachs CEO David Solomon flagged concerns about the "unchecked advancement of frontier models," saying: "We must enable innovation and also demand responsibility."
- Wells Fargo CEO Charlie Scharf said the bank is evaluating third-party models and agents while "considering the potential risk these tools can introduce."
Meanwhile, the banks build their own agents
The caution is reserved for agents built by someone else. JPMorgan says it already runs agentic AI inside roughly 80 internal workflows. And on the same earnings call, CFO Jeremy Barnum disclosed that the bank recently began a targeted pilot of its own AI tool, Smart Cash, which predicts a customer's cash flow and moves funds to maximize yield.
That is the competitive subtext: JPMorgan is blocking external agents from doing exactly what its own agent will do — inside the bank's security perimeter, under its compliance controls, keeping the customer relationship and the deposit income at home.
What it means for founders and investors
Three implications for the agentic-commerce buildout:
- Distribution runs through the banks, not around them. If the largest banks wall off customer accounts, personal-finance agents lose access to a large share of the market. Startups need bank partnerships and verified-identity frameworks, not just better models.
- Open banking beats screen scraping. Read-only connectors are the permitted lane; credential-based access is the blocked one. Product architectures should assume this split persists.
- Compliance is a moat. Governance platforms like Citi's ARC show where the industry is heading: identity, traceability, and human accountability as prerequisites for execution. Startups that build to that standard from day one have a head start on incumbents' procurement checklists.
Angeled.com connects founders with investors.
Sources: reported October 16, 2026 — Startup Fortune; Crypto Briefing, via the Financial Times.
Read Next
OpenAI Releases Codex Security as Open-Source CLI and SDK
Databricks Unveils Always-On Genie, an AI Agent That Stops Waiting to Be Asked
DeepMind Launches AlphaGenome 2, a DNA Model That Reads Genomes Cell by Cell
ElevenLabs Unveils 13 New Products and Upgrades at Its New York Summit
Google Launches Gemini Decision 1, a Model That Answers Instead of Writing
Runway Open-Sources Praxis-1, Its Robot Model That Learned by Watching Video



