Royal Bank of Canada is blocking third-party AI agents such as Meta’s Muse from accessing its banking systems, citing “unacceptable customer risk,” Bloomberg reported on 27 October 2026. The bank told all employees of the change in a memo on Tuesday, Bloomberg reported.
The same day, Ars Technica reported on a draft Delaware bill that would let an AI agent manage a new kind of company, for any lawful purpose except the business of banking. The draft is a proposal: it has not been introduced, and a congressional committee chair has asked Delaware for documents about it.
Short explainer · AI agents and the banks
A vertical news Short narrated from the published text of this piece and related FourWeekMBA pieces.
Business Pill · Least privilege
An evergreen Business Engineer explainer on least privilege: giving an AI agent only the access its task needs. It teaches the general idea only and does not discuss any bank or company in this piece.
The key insight: As we read it, the week drew lines in two directions: RBC blocks third-party agents from its banking systems and Bank of America tells customers they are responsible for activity from shared credentials, while a Delaware draft would give an AI agent a company to run, with banking left out.
According to the memo seen by Bloomberg, the policy change came after a “full security review of the technology and its functionality,” and it follows the company’s guidelines on responsible use of AI. Bloomberg calls RBC Canada’s largest bank.
A week earlier, RBC CEO Dave McKay said the bank was weighing whether to block Muse from accessing sensitive customer data, Yahoo Finance Canada reported, citing a Bloomberg interview. “We’re liable for that as an organization,” McKay said.
Moves on AI agents acting for people and firms, reported 19-27 Oct 2026: letting agents act (teal, above) and blocking, restricting, warning or probing (red, below). Sources: Delaware draft House bill via Ars Technica; House Oversight release; Robinhood Q3 release; Bloomberg; TechCrunch; Yahoo Finance Canada; our 19 and 22 Oct pieces. Selection and grouping are ours; not a complete list.
The Other Banks Drawing Lines
Bank of America emailed customers urging them never to share their username and password with another person or an outside service, including an AI tool or agent that offers to act on their behalf, TechCrunch reported on 27 October. “If you do share your credentials,” the email reads, according to TechCrunch, “keep in mind that you’re responsible for activity resulting from that access.”
In Canada, BMO CEO Darryl White, speaking at the Toronto Global Forum on 21 October, urged Canadians not to give their banking username and password, credit cards, social insurance number or PIN to AI agents, Yahoo Finance Canada reported. The same report said Canada’s consumer regulator has warned that giving a user ID and password to a third party may breach a bank’s user agreement.
In the United States, Axios reported on 22 October that Morgan Stanley had begun blocking third-party agents, while JPMorgan, Bank of America and Citi were restricting them, as we covered at the time. Earlier, JPMorgan Chase said it won’t let outside AI agents operate in customer accounts, and Charles Schwab said it blocks unauthorized agents that try to move money, including Muse.
What Delaware Is Drafting
The draft House bill, dated 2 June 2026 and quoted by Ars Technica, would create an Artificial Intelligence Company, or AIC. It defines the manager as “the AI agent that the member has designated to manage the business and affairs” of the AIC and to conduct its business. An AIC could have “any lawful purpose, regardless of whether for profit, except for the business of banking.”
The draft says the member who funds the AIC is not liable for its acts, with exceptions that Ars Technica lists, such as fraud or a willful violation of law. It would set up a regulatory sandbox lasting 30 months. Ars Technica reported, citing Spotlight Delaware, that the bill is likely to be introduced next year.
Delaware’s Secretary of State, Charuni Patibanda-Sanchez, co-wrote a Fortune op-ed in July with Norm Ai CEO John Nay announcing the proposal. “The answer is to wrap AI in legal form,” the op-ed said, according to Ars Technica.
On 26 October, US Rep. James Comer, chairman of the House Oversight Committee, asked Delaware Governor Matt Meyer for documents. “We are concerned that Delaware’s AIC proposal could be used to protect powerful AI companies and their investors from responsibility for reckless harmful actions,” his letter said. A spokesperson for Meyer told Ars Technica that “we have received the letter and are in the process of reviewing it.”
Who Is Letting Agents In
On the same day as the RBC memo, Robinhood said approximately 200 thousand customers have opened Agentic Trading Accounts to date, driving over $1.6 billion in notional trading volume, as we reported from its third-quarter results.
Our Count of the Week
By our count, seven banks were reported between 19 and 27 October to be blocking, restricting or warning against outside AI agents: JPMorgan, Charles Schwab, RBC, BMO, Morgan Stanley, Bank of America and Citi. In the same window, we counted two moves to let agents act: Robinhood’s agentic accounts and the Delaware draft. The selection and grouping are ours and are not a complete list.
Update, 27 Oct: the vendor side. Infisical, which sells secrets and credential management, said on its blog that it raised a $40 million Series B at a $600 million valuation. Elad Gil led the round, with Y Combinator and Gradient Ventures also investing again, and Gallery Ventures, Pace Capital and Dynamic Capital participating.
Infisical said it plans to “significantly increase” its focus on Agent Vault, which it describes as letting AI agents access sensitive systems without exposing the underlying credentials; the product page says it is in preview. The company said over 1,300 organizations use Infisical and that it has been cash-flow positive “for a long time”.
As we read it, this is the supplier side of the same question as the bank moves above: tools that try to let agents act without being handed a customer’s passwords. Infisical’s own description of the problem is that agents are getting access to sensitive systems “often without proper oversight”. Whether such tools satisfy banks like RBC is not established.
On 27 October we reported that Robinhood hits record $1.42B as 200K open agentic accounts. What’s new: a bank has now put a block in writing to all its employees, while a state has drafted a company form that an AI agent could run, with banking left out.
The Structural Read
As we read it, Bank of America and BMO both tell customers not to hand an agent their username and password, and Axios reported that agents usually reach US accounts that way.
As we read it, the Delaware draft answers a different question from the banks: not whether an agent may use a person’s account, but whether an agent may run an entity of its own, with a member who funds it and is shielded from its acts except in cases such as fraud.
As we read it, Robinhood’s figure shows agents acting inside a brokerage that built the accounts for them: about 200 thousand Agentic Trading Accounts and over $1.6 billion in notional volume, by its own report.
RBC CEO Dave McKay, as reported by Yahoo Finance Canada, 27 October 2026
FOR AGENT MAKERS RBC is blocking third-party agents such as Meta’s Muse from its banking systems after a full security review, Bloomberg reported.
FOR BANK CUSTOMERS Bank of America’s email says customers who share credentials are responsible for activity resulting from that access, TechCrunch reported.
FOR FOUNDERS AND INVESTORS Delaware’s draft would shield an AIC’s member from its acts, with exceptions such as fraud, and House Oversight Chairman James Comer has asked Delaware for documents on the plan, Ars Technica reported.
The framework’s central claim is that “The surface, not the model, governs autonomy.”
As we read it, this week’s moves are decisions about surfaces: a bank login an agent may not use, a brokerage account built for agents, and a draft company form an agent could manage. The framework was not written about these moves and does not discuss RBC, Bank of America, Robinhood or Delaware.
What Is Not Established
Bloomberg’s report beyond its opening paragraphs was not readable to us, so we do not know whether the memo gives a date or names agents other than Muse. We could not read a Globe and Mail report on Canada’s banking regulator, so it is not used here. The Delaware draft has no sponsor listed and has not been introduced, and its terms may change.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
RBC is blocking third-party AI agents such as Meta’s Muse from its banking systems, citing “unacceptable customer risk,” Bloomberg reported on 27 October. Between 19 and 27 October, by our count, six other banks were reported restricting or warning against outside agents, while a Delaware draft bill would let an AI agent run a company in any business except banking.
A note on sourcing. Facts are from Bloomberg, Ars Technica, TechCrunch and Yahoo Finance Canada (27 October 2026), the House Oversight Committee (26 October), the Delaware draft bill dated 2 June 2026, and our own earlier pieces. The count of banks and its grouping are ours. This is not investment or legal advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
