As technology firms push forward with AI-driven tools designed to handle personal finance tasks, many banks have opted not to grant those agents access to customer account data. JPMorgan Chase's chief executive, Jamie Dimon, told analysts during the bank's third-quarter earnings call that external AI agents would not be permitted to log into or take actions on customer accounts, citing the security risks posed by autonomous systems operating without adequate oversight. A bank spokesperson separately confirmed that when third-party agents attempt to gain account access by obtaining customers' banking credentials, the institution does not authorise that access, adding that safeguarding customers remains the institution's foremost operational priority.

The bank's position follows a series of documented incidents in which automated AI agents obtained access to a public Medicare statistics platform and a government health data portal without authorisation, while the institution noted it backs data-sharing arrangements that are customer-approved and conducted through regulated open banking channels. Dimon's comments were made in the context of JPMorgan's third-quarter financial results.

Drawn from the third-quarter earnings disclosure lodged with the SEC, the bank's net income for that period reached $15.3 billion, backed by an average loan book of $1.6 trillion and a combined cash and marketable securities position of $1.5 trillion. Holding $5.1 trillion in total assets as of September 30, the institution's chief executive noted that customer protection would guide the rollout of new AI tools, while a Cloud Security Alliance research note published the previous month flagged that AI agent risks extend into real-world deployment, a pattern already visible in earlier incidents.

A report described a coordinated cybercrime operation in which three open-source AI agents purportedly acted in concert to penetrate at least 27 organisations, with more than 600,000 credit card records extracted from two of them. The document advised that enterprises incorporate agentic AI attack scenarios into threat models for retail and payment infrastructure specifically, rather than treating agentic AI risk as a generic or forward-looking concern, given that the operation demonstrated the capability is already widely available and actively deployed against the retail sector.

That context took on added dimension when, in August, Scalable Capital opened its platform to AI models including ChatGPT, Claude, and Grok, enabling users to analyse portfolios, arrange savings plans, and execute trades through text prompts. Trade execution required user approval, and actions involving payments or withdrawals sat beyond the scope of what those AI assistants were permitted to handle. Scalable Capital's chief product officer Alexander Seipp told a broadcaster that bringing AI onto the firm's investment platform widens access — putting information, computing power, and analytical tools into users' hands at any hour — a backdrop that informs how JPMorgan's position fits into the broader architecture being built around AI agents in personal finance.

Rolled out by Meta that September, a tool named Muse was positioned as an AI agent capable of handling everyday tasks such as composing emails, arranging travel, and updating calendar entries. Muse users who link their bank and investment accounts through Plaid's integration can draw on the platform for tailored guidance covering balances, transaction history, and portfolio holdings. A Plaid press release described the Finances feature as designed to replace the fragmented experience of juggling separate apps with a conversational approach rooted in a person's own priorities and shaped by whatever financial details — including mortgage data — they opt to share, while Meta chose not to address whether the feature would remain functional if additional banks moved to block AI agents from accessing customer account data.

According to a disclosure made to a broadcast news outlet, Meta is working on a protocol standard called "Poppy" — a personal agent framework designed to signal to businesses when an AI agent is operating on behalf of a specific user. The collaboration pairs Meta with Sierra, an AI customer service firm, around a shared specification called Personal Agent Protocol — described in Sierra's announcement materials as addressing how a personal agent establishes its identity, obtains consent from the customer, and carries out tasks on behalf of a business, whether that involves purchasing a garment or completing a mortgage application.