Analysis
JPMorgan Chase won't let outside AI agents, including ones like Meta's Muse, log into or act on customer accounts, CEO Jamie Dimon said on the bank's third-quarter earnings call Tuesday, warning that such tools create the risk of "rogue agents" operating at scale. Fortune and Banking Dive both reported Dimon's remarks, which land as Meta's Muse, launched in September with the ability to connect to financial accounts through Plaid, pushes further into consumer finance. A JPMorgan spokesperson confirmed the bank currently blocks external AI agents from acting on behalf of retail customers, even when a connector like Plaid has the user's login credentials.
A liability question banks haven't settled
Dimon's framing centers on who bears the risk: customers who hand an AI agent their banking credentials are making that choice and own the consequences, he said, calling it a "liability transfer." He drew a sharp distinction between read-only access, where agents can already see balances and other account data through Plaid-style connectors, and the ability to actually move money, which JPMorgan still reserves for the bank's own tools, including an AI-powered cash-management product called Smart Cash that's currently in pilot. Banking regulators are paying attention without yet writing a rule: the Federal Reserve, OCC and FDIC have started asking banks how they're thinking about agentic AI risk to deposits and liquidity management, gathering information rather than mandating a specific approach.
“The remarks landed inside a broader earnings call where JPMorgan's own AI economics are still murky.”
Not every bank is taking JPMorgan's position. Germany's Scalable Capital opened its platform in August to AI assistants including ChatGPT, Claude and Grok, letting them propose trades that a human still has to approve, with no ability to withdraw funds directly. That's a materially different bet than JPMorgan's outright block, and it gives the market a live comparison of which approach customers and regulators prefer once something inevitably goes wrong with one of them. Meta, for its part, declined to say whether Muse could still offer financial assistance if more banks follow JPMorgan's lead; it's instead developing a separate "Poppy" personal-agent protocol with Sierra, which Sierra describes as complementary to its own agent framework rather than a replacement.
The remarks landed inside a broader earnings call where JPMorgan's own AI economics are still murky. CFO Jeremy Barnum put the bank's annual AI spending near $700 million, a figure he immediately called "a little stale," against Dimon's 2025 comment that AI-driven savings had roughly matched spending at the time.
Quarterly expenses rose 12% year-over-year to $27.3 billion, with the bank guiding to about $27.5 billion in the fourth quarter, and Barnum said AI's benefits aren't yet clearly visible in the numbers. What the headline policy decision misses is that JPMorgan isn't rejecting agentic AI, it's trying to keep the agent layer in-house while outside players like Meta figure out distribution first. For fintechs and AI-agent startups selling into banking, the practical read is that the fastest path to a bank's customers right now runs through partnership with the bank's own tooling, not around it.
