AI Agents With Login Access Reach Bank Accounts

Security experts warn consumer agents holding full credential vaults are already acting on finances, with no settled safeguards in place.

Updated on Nov. 2, 2026 in Artificial Intelligence

Bold vector editorial illustration of a steel strongbox cracked open to spill brass keys across a slate surface, evoking unguarded access to personal finances.
Security experts warn that consumer AI agents already hold full credential vaults and act on customer bank accounts with no settled safeguards in place. AI Illustration. Upload story photo >

Live Poll

Would you give an AI agent app your bank login to manage your finances?

Alex Stamos, chief security officer at AI coding startup Cognition and former Meta security chief, called giving AI agents access to personal financial logins "totally insane." The concern centers on consumer agents that users authorize to manage finances or file taxes, a practice that is already in use rather than a future product.

Why it matters

Consumer AI agents are moving from chat interfaces into logged-in accounts, where they act on users' behalf with real money and records at stake. Banks now see customer-authorized bot activity that is authenticated correctly, which makes it harder to distinguish legitimate delegation from abuse.

No benchmark or performance figures were reported for the agents involved. The only quantified detail is that the Instinct company is six months old, according to Stamos.

The players

Alex Stamos

Chief security officer at AI coding startup Cognition and former Meta security chief.

Jamie Dimon

CEO of JPMorgan Chase, which is developing its own agentic tools for clients.

Meta

Social media company that launched its Muse AI agent in September, running on its own cloud computer.

The details

Users give AI agent apps their entire 1Password vaults, the encrypted stores of passwords and login credentials, and ask the agents to handle finances or file taxes. Banks see these interactions as customer-authorized bot activity that is authenticated correctly, not stolen credentials, which is why standard fraud checks may not flag them. Stamos said Instinct is popular in part because it bypasses safeguards that are meant to prevent bots from operating in these cases.

Timeline

  1. September: Meta launched its Muse AI agent, which runs on Meta's own cloud computer, according to a Meta blog post.

  2. October: JPMorgan Chase CEO Jamie Dimon warned consumers about granting third-party AI agents access to financial accounts during an earnings call.

  3. November 2, 2026: The article was published, with Stamos's comments made on the Big Technology Podcast.

The Tech Race

Meta's Muse agent, launched in September, runs on Meta's own cloud computer, placing it among the consumer agent products now competing for logged-in access to users' accounts. Instinct, a company six months old, is an example of a smaller entrant gaining popularity by bypassing safeguards that block bots from these tasks.

Users who authorize AI agents to manage finances or file taxes are giving those agents their full credential vaults, and banks may treat the resulting activity as legitimate customer behavior. Availability and any safeguards for these agents were not specified in the reporting.

The takeaway

Consumer agents that hold full credential vaults are now active, and banks are seeing their authenticated activity as normal customer behavior. Watch whether banks such as JPMorgan adjust fraud detection for customer-authorized bot traffic as their own agentic tools for clients develop.

Further reading

For more on the agents now reaching into personal accounts, see the Artificial Intelligence section.

Source note: This article includes information reported by Business Insider.

Live Poll

Would you give an AI agent app your bank login to manage your finances?