If you have been letting an AI helper sign in to your Royal Bank account to review spending or move money around, that door is closing. Canada's largest bank has decided that outside personal AI agents may no longer connect directly to its authenticated digital services, online banking included, and links customers already set up will stop working.

The decision, published on October 27, makes RBC the first Canadian bank to draw this line, according to reporting by The Globe and Mail — and it lands in the middle of a much wider argument about who answers when an autonomous program acts with your money.

What changed#

The scope of the block is specific. It covers third-party personal agents reaching RBC accounts through authenticated channels — assistants that log in as you to check balances, analyse transactions or take actions on your behalf. It does not close your account, change your login, or touch the bank's own app and website, which keep working exactly as before.

RBC has not published a list of named products caught by the policy. Coverage has singled out Meta's Muse as the prominent example — notable, since chief executive Dave McKay had said on Bloomberg Television a week earlier that the bank was weighing a block on Muse specifically. A same-day staff message from technology and operations chief Naim Kazmi and personal banking chief Erica Nielsen said blocking work would begin immediately, and spokesperson Jeff Lanthier confirmed the policy.

Why RBC says it had to act#

The bank's public reasoning is simple. To do anything useful, a personal agent needs the keys to your accounts, and it may store your passwords to get there. While it is merely reading a statement, that is one level of exposure. Once it can move money or act across services on its own, a single misread instruction stops being an annoyance and becomes a loss.

RBC's notice puts it in terms of escalation: as these systems become more autonomous, the stakes of something going wrong climb from fraud and consumer harm to risks that could ripple through the financial system. The internal message was blunter, describing "unacceptable customer risk" and warning that agents can act beyond what customers expect, leaving the bank carrying liability for actions it never approved — liability no one has yet settled, in law or in contract, for the case where the agent, not the customer, pressed the button.

A smartphone beside a locked digital gate in front of a grand bank building
AI-generated editorial illustration.

A door closing across North America#

RBC is not acting in isolation. It is the Canadian chapter of a story building all month on both sides of the border:

  • United States, mid-October: PYMNTS reported that four of the biggest US banks were turning agents away: Schwab blocking unauthorised agents, JPMorganChase's Jamie Dimon saying outside agents would not get in, Bank of America limiting agent access while it writes liability rules, and Citigroup not letting third-party agents log in or transact.
  • Canada, October 28: Bloomberg reported that Bank of Montreal, Toronto-Dominion Bank and Bank of Nova Scotia were also moving to restrict third-party agents. BMO's chief executive has separately warned Canadians against handing agents their banking logins, PINs, card numbers or social insurance numbers.
  • The regulator: the Office of the Superintendent of Financial Institutions has warned that agentic AI introduces cybersecurity and operational risks, but is reportedly leaving each bank to decide rather than ordering a blanket ban.
  • The earlier precedent: in September, Amazon locked Meta's Muse out of its storefront. Different industry, same logic: the owner of an authenticated system gets to decide which outside agents are allowed inside it.

The pattern is clear: the institutions holding the accounts are setting the default to "no" and inviting agent platforms to negotiate their way back in on the institutions' terms.

A robotic hand reaching toward a closed bank vault door marked with a shield
AI-generated editorial illustration.

What this means for the agent economy#

The agent industry's pitch is that it moves past answering questions into doing things. Banking is the hardest test of that pitch, because money is irreversible, heavily regulated and deeply personal.

For agent makers, the implication is a fork in the road: keep reaching accounts via credential-sharing tools and watch the doors close one bank at a time, or negotiate official, permissioned access — scoped APIs, delegated authority a customer can revoke, and liability terms agreed in advance. For customers, the point is simpler: you can still get AI help with budgeting without handing an agent your bank login, and the bank's own tools are unaffected. A January PYMNTS Intelligence report put it well: broad adoption needs "payments-grade" trust — a bar RBC says today's agents do not meet.

What to watch#

Three questions will decide whether this is a pause or a permanent settlement. First, will RBC or any peer publish criteria for approved agent access? A published standard would turn a blockade into a market. Second, will the restrictions converge, or will some banks compete by offering controlled agent access as a feature? Third, will regulators bless a standard for delegated agent authority, as open banking did for data access, or will each bank keep improvising its own perimeter?

For now, the message from Canada's biggest bank is blunt: your agent can advise you about your money, but it cannot be you at the bank. The agent economy will have to earn its way back inside — with permission, with limits, and with someone clearly responsible when things go wrong.

Sources: RBC's customer notice as reported by PYMNTS (October 28, 2026); the staff memo and spokesperson confirmation as reported by Briefs, citing Globe and Mail and Bloomberg reporting; Money.ca on OSFI's warning; and Bloomberg on peer-bank restrictions. Cover photo: public domain via Wikimedia Commons; inline illustrations AI-generated.