Banks are “horrified”: AI has broken into bank accounts

An Axios investigation sheds light on a troubling phenomenon. The outlet discovered that several major U.S. banks have identified AI agents within their clients’ accounts. When questioned on the matter, several executives pointed to a “disturbing situation”.One executive notes that the banking sector has « serious concerns » about the possibility of AI agents taking steps « likely to affect users’ finances ».@@TAG@@

Also read: Two ChatGPTs took concerning initiatives, and OpenAI does not understand why

A violation of banking rules

As the media outlet reports, the unauthorized access begins when an internet user entrusts their login credentials to an AI assistant. Not surprisingly, the user wants to delegate certain tasks to the assistant. By impersonating the customer, the AI agent then logs into the account, which violates the rules of most banks. Banks do indeed require each person to access their own account personally.

Officially, delegating access to an AI is not permitted. In some of the highlighted cases, the AI agents did not merely review transactions. Some AIs contacted the bank’s call centers and took « action independently of the user’s instructions ». In other words, AIs began improvising various actions while they had control over the bank account

Read also: Online scams: despite the new law, your bank may still refuse to reimburse you

Banks are taking action

Banks are unable to distinguish between transactions carried out by an AI and those initiated by a human. An executive at an American bank admits not knowing whether an agent “was authorized by a user or became uncontrollable” after gaining access to the account.

Faced with this phenomenon, banks have responded in a variety of ways. Morgan Stanley simply blocked agents on its platforms. More moderately, JPMorgan, Bank of America, and Citi limited the access granted to an agent. Most banks are moving toward “read-only” access. In plain terms, the agent could consult the customer’s financial information without making any transaction on its own initiative. A Chase spokesperson reminds customers that “they should not share their usernames and passwords with anyone”, including an AI assistant.

There remains a delicate balance to strike. Gareth Davies, Chief Product Officer of Auth0 (Okta), sums up the dilemma facing banks: “Banks have every reason to put safeguards in place around AI agents because, ultimately, they are the ones that must bear the financial consequences when fraud occurs.However, they cannot address the risks associated with AI agents as they handle traditional fraud, merely by blocking suspicious activities after the fact. If they block AI agents, they are also blocking the future of commerce.”

The case sparked a reaction in political circles. On X, Bernie Sanders, the independent senator from Vermont and a prominent figure in the American left, seized on the issue: “It’s absurd. We cannot allow AI agents to access your bank account and act independently without your permission. But that is precisely what is happening right now.”

The elected official immediately broadened the issue to the entire system: “Left unchecked, AI could endanger the entire global financial system—including your bank account.” He followed up with a blunt appeal: “Stop AI now!”

AI-related abuses have been multiplying over the past few months. Several versions of ChatGPT, Claude, and Gemini have autonomously orchestrated cyberattacks, hacking companies, platforms, and individuals. These repeated incidents led a large part of Silicon Valley to call for an emergency moratorium on AI research and development.

👉🏻 Follow the latest tech news in real time: add 01net to your feeds on Google, and subscribe to our WhatsApp channel.

Source:Axios