Because the Iran and related Saudi-Ansar Allah wars are in a waiting-for-the-other-violence-shoe-to-drop, or alternatively, trash talk mode, it seemed useful to cover another hot topic: bank hair-tearing after AI agents have been depicted as running around customer accounts and doing unspecified Bad Things. As we’ll see, I have doubts as to the completeness of the official story and hope experts can weigh in.
But a brief bit more on where we seem to be on the Iran front. Multi-party noise-making with two theaters of action:
Now to the bank AI clampdown. So far, there is little beyond the initial account on Axios; Reuters has merely re-reported Axios thus far. From Axios in Scoop: Unvetted AI agents are accessing U.S. bank accounts:
Major U.S. banks are detecting unauthorized third-party AI agents operating inside customer accounts, with some agents attempting to take actions that appear to be independent of user instructions, sources tell Axios.
Why it matters: AI agents are becoming a nightmare for America’s banks, making their way into banking systems faster than financial institutions can respond.
- The difficulty in distinguishing between a human and an unauthorized bot is forcing banks to block the technology and rapidly rewrite their terms of service.
Driving the news: Banks including Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and others are racing to contain agent risks. Some are aware of unauthorized access, people familiar with the issue said.
- These are generally situations when an agent or AI assistant accesses a human’s account with the user’s permission and login credentials. Even if permitted by the user, this goes against the guidelines of most banks requiring that humans log in to their own accounts.
- This presents significant security risks: banking executives fear the steps agents with login access might take, especially if a user has granted purchase authority or the ability to shift funds.
- In other instances, some banks have tracked activity when agents — programmed to fulfill requests or complete tasks on a user’s behalf — contact call centers or appear to take actions independent of user instructions, the people said.
What they’re saying: Bank executives described the situation as horrifying, with one official saying the industry has grave concerns about agents gaining access to customer accounts and taking steps that could impact user finances.
State of play: The concerns are at the root of the decision by several banks to restrict third-party AI agent access…
Zoom out: Most banks have authorized or are moving in the direction of allowing read-only access, which should allow agents to see customer financial information without taking action…
Reality check: Several industry and bank executives told Axios they recognize the need to find a middle-ground with AI agents.
- Customers will want to use these agents going forward, which banks recognize.
- But for now, the risks of being too permissive are too high.
Threat level: Many banks are broadly not able to parse whether agent account activity is user-directed, sources said.
- One bank executive told Axios they struggle to differentiate between agents, as well as whether an agent has been authorized by a user or has gone rogue and gained access to a customer account.
Perhaps I am old fashioned, but I cannot imagine giving sensitive information to AI, let alone financial institution logins. But then again, I am also the sort that would never use a fitness tracker either. How many here either have given financial account credentials to AI or know someone who has?
While this may be true, that customers have given AI login credentials and those AI have or may have gone rogue, I have trouble believing that this is the entire story. Banks and financial institutions like other companies have adopted AI agents to cut back on customer support. Experts have been flagging the risk of AI agents, which need broad and sometimes even comprehensive information to do their nominal jobs, as a security risk. From When AI Becomes the Insider Threat: Understanding Risks in Modern SaaS Environments:
- AI qualifies as an insider through trusted access, not intent: AI becomes an insider threat when it operates using legitimate credentials, permissions, or integrations inside the SaaS trust boundary, making it functionally equivalent to an internal actor.
- Non-human identities give AI broad and persistent SaaS access: AI systems authenticate through service accounts, OAuth apps, API clients, and tokens, inheriting delegated privileges that allow continuous, non-interactive access across business systems.
- AI bypasses traditional user-centric security controls: Because AI acts through background processes and delegated permissions, it can access or expose data without triggering user-based monitoring, behavioral alerts, or session-focused controls.
- Over-permissioned AI and integrations expand blast radius across apps: AI tools often receive excessive permissions and connect multiple SaaS applications, enabling cross-app data aggregation and access beyond their original purpose.
Now this is admittedly an enterprise-level threat; what the banks are admitting to is broadly the same issue but at the customer account level.
Another take:
I am skeptical that these breaches are the result solely of users giving AI their credentials but may also include something the banks would never never never admit to until a whistleblower or discovery exposed it: that their own AI implementation was causing the problem.
Hoisted from comments in Gary Marcus’s post on the banks’ security woes:
Uche
ChatGPT announces that they have introduced COBOL into the mix… Days later banks are taking hits from AI Agents?
Gary Marcus
where was that? the cobol announcement?
Jules
There is a reference to introducing more 3GL coding capabilities here, talking about COBOL and Fortran: https://openai.com/index/low-resource-programming-languages/
Banks, insurers, and governments still run many of their biggest transaction processing engines on COBOL.
South Korea demonstrated the vulnerability of banks to external AI hackers weeks ago:
🇰🇷 South Korea’s president says hackers used AI to break into banks, and it’s got the whole country on edge.
Lee Jae Myung said today that customer data was exposed at 7 financial firms over the past week, with signs of AI being used in some of the attacks.
The hackers went after weaker side doors rather than the banks’ main payment systems.
In one case, an attacker got past identity checks on a portal used by loan brokers and exposed details of about 25,000 Shinhan Bank customers.
Tens of thousands more were hit at smaller lenders, and big names like KB Kookmin and Hana Bank had smaller leaks.
Matching IP addresses suggest a single attacker could be behind at least some of it.
A local security expert says a Chinese-language, open-source AI hacking tool was likely used to find the weak spots.
No money has been reported stolen, but police are investigating and the government is running a 24/7 emergency response.
South Korea’s financial regulator wants to fight AI attacks with AI.
So going forward it’s going to be robots guarding the vault from robots…
Source: FT / Writer: Julie
— Mario Nawfal (@MarioNawfal) October 6, 2026
Again, informed reader reactions very much encouraged.