Banks Block AI Agents That Log In as You: Lessons for Creators [2026]

Banks Block AI Agents That Log In as You: Lessons for Creators

In mid-October 2026, two of the biggest US financial firms told AI agents to stop signing in with customer passwords. JPMorgan Chase and Charles Schwab both said outside agents may not log in on a customer's behalf, according to coverage of their earnings calls. Creators and brand teams are doing the opposite: handing agents the keys to Instagram, TikTok, YouTube and ad accounts, and hoping nothing goes wrong. This guide explains what the banks decided, why it matters for anyone who earns a living on a social account, and the rules to copy before an agent touches yours.

At a glance

Question	Short answer
What did the banks do?	Blocked outside AI agents that sign in with a customer's login; kept read-only routes open
Why?	An agent with your password looks like you, so nobody can tell who acted or who is liable
Why should creators care?	A social account is your storefront, and a login lets an agent change the recovery email as well as post
Is there a real example?	Meta says 20,225 Instagram accounts were taken over through a flaw in its AI-assisted recovery tool
What to do now	Use scoped access, two-factor authentication, team roles and a human approval step for money and recovery changes

What exactly did JPMorgan and Schwab decide?

JPMorgan CEO Jamie Dimon said on the bank's earnings call that the bank would not let outside agents operate inside its systems, and warned about "rogue agents" that could number in the thousands, according to Cybernews. A JPMorgan spokeswoman said third parties cannot use customers' banking credentials to log in for them.

Schwab CEO Rick Wurster said the firm blocks agents that try to sign in for clients and stops dozens of such attempts each week, American Banker reported. Schwab points clients to its own assistant instead.

The liability point is the one creators should read twice. JPMorgan CFO Jeremy Barnum called credential sharing insecure and said customers who give an agent their logins could be responsible if it acts badly, American Banker reported.

The restriction is narrow. Read-only access is still available through approved connectors, and the banks are blocking agents that pretend to be the customer, not AI features in general.

Why does a password make an agent so risky?

A password carries no purpose. When an agent signs in with your credentials, the service sees you, with every power you have. It cannot tell "schedule this post" from "change the recovery email and lock the owner out".

That is the whole argument in one sentence. The banks are not afraid that agents are clumsy. They are saying that a login is the wrong shape of permission, because it cannot be limited to a task, cannot be told apart from the human, and is hard to revoke without changing the password everywhere.

Has an AI-assisted account system actually been abused on Instagram?

Yes, though on the platform's side rather than the user's. Meta says attackers hijacked 20,225 Instagram accounts by exploiting a flaw in its AI-assisted recovery system, High Touch Support, Help Net Security reported on June 8, 2026.

Here is how the reported attack worked:

1. Attackers used VPN services to appear in the same general area as the target.
2. They asked Meta's AI support assistant to link the account to an email address they controlled.
3. The system sent a password reset link to that address.
4. They reset the password and logged in.

Meta's Associate General Counsel for incident response, Amber Hannah, said the tool "functioned as intended" and blamed a bug in a separate code path that failed to check that the requester's email matched the account. A Maine Attorney General filing lists April 17 as the incident date, and Meta found the problem on May 31.

Accounts with two-factor authentication were targeted but not compromised. Engadget first reported that they were exposed and later added a correction saying they were not taken over. Reported victims included the dormant Obama White House account, and Meta VP of communications Andy Stone wrote on X: "This issue has been resolved and we are securing impacted accounts."

The lesson is the same as the banks': the more power an automated helper has over account recovery, the more an attacker gains by talking to it. For a creator, the account is the business. If you want the background on how brands vet creators, our guide on how to check fake followers covers what a healthy account looks like.

What do creators and brand teams usually give agents today?

In practice, three things, in increasing order of risk:

• Drafting help. The agent writes captions or scripts. No account access. Low risk.
• Publishing and scheduling. The agent posts through a connected tool. Medium risk if the tool uses a limited token, high if it uses a stored password.
• Inbox and account management. The agent answers DMs, accepts collaboration requests or touches settings. High risk, because inbox content is untrusted text that someone else wrote.

Brand teams often run the second and third at scale. If you manage many creators, the same rules apply to each handle, and our influencer management guide is a good place to set up roles before you automate anything.

What can you copy from the banks?

Reported options at the banks are worth borrowing as rules:

1. Separate read access from acting as the owner. Let an agent view analytics. Do not let it log in as you.
2. Use fixed connectors with a short list of tools. A scheduling connector that can publish drafts is safer than a browser session that can click anything.
3. Give the agent its own identity. A named team member or token can be removed in one click and shows up in the activity log.
4. Keep a human in the loop for money and recovery. Payout details, linked emails, phone numbers and password changes always need a person.
5. Assume untrusted text will try to steer it. Anything the agent reads in a DM, comment or email can contain instructions. Do not let that text trigger account changes.

How do you lock down a creator account this week?

1. Turn on two-factor authentication with an authenticator app, not SMS.
2. Check the recovery email and phone number. Make sure only you control that mailbox.
3. Open the list of devices and connected apps. Remove anything you do not recognise.
4. Move team members to roles, not shared logins. Our guide to Instagram influencer outreach explains how teams usually split the work.
5. If an agent already posts for you, find out whether it uses a token or your password. Replace passwords with tokens.
6. Write down who can approve payouts and brand deals. Agents do not approve either.
7. Re-check after every collaboration where a brand asked for temporary access.

How can you see what a coding or marketing agent is doing on your machine?

Not every agent lives inside a social app. Many creators and small teams now build their own tools with AI coding agents: a link page, a media-kit generator, a script that pulls analytics. Those agents run on a laptop that also holds the logins to the accounts above.

AgentBeam is a security and observability tool for AI coding agents from our team. It runs locally, flags credential exposure, redacts secrets before storage, and can deny specific actions through Claude Code and Codex hooks, per its own write-up of the bank story. It does not sit between browser agents and a bank or a social platform, and it cannot enforce a platform's policy. It is useful if you run coding agents next to your creator accounts and want to see what they read and run.

Is the ban only about safety?

Coverage of the bank decision mixes safety and business motives. Critics quoted in the AgentBeam analysis note that data-access fees and customer revenue may also play a part, and that blocking is leaky because users can still run browser agents themselves. Treat the banks' stance as a signal about permission design, not as proof that every agent is unsafe.

What should you do now?

Pick one account, the one that earns the most, and apply the six steps above before the end of the week. Then write a one-page rule for your team: agents get connectors and tokens, never logins; people approve recovery changes and money. That single page will do more for your account security than any new tool.

Conclusion

The banks did not ban AI. They banned agents that act as the customer, because a login cannot carry a limit. Creators face the same trade-off with much less protection: a hijacked Instagram account costs reach, brand deals and income, and the Meta incident showed how quickly automated account help can be turned against its users. Keep agents on scoped, revocable access, keep humans on recovery and money, and review your connected apps today. For more on building a safe creator workflow, start with our guide to Instagram influencer marketing and our Instagram collaboration guide.

Details in this article are accurate as of October 25, 2026. Check each source for later changes.