Big Banks Answer the AI Alarm: Humans Stay in the Loop

Big Banks Answer the AI Alarm: Humans Stay in the Loop

Bank of America's Brian Moynihan calls AI companies' focus on safety "encouraging" — and assures that the bank never lets AI agents operate alone. PNC says agent control is a top priority. The figures behind the assurances are unverified company numbers.

When Bank Leaders Answer Their Industry's Own Alarm

On Monday, September 14, 2026, at the Barclays Global Financial Services Conference, the leaders of two of America's largest banks used words few expected from the banks betting hardest on AI themselves: caution.

Bank of America CEO Brian Moynihan called it "encouraging" that AI companies are taking responsibility for AI safety seriously, according to Bloomberg (d1404d4a). At the conference he was more specific about why:

"This weekend was a lot about the risk in AI, and that largely is around agents just left to operate," Moynihan said, according to American Banker (3937add3). "We just don't do that."

The Week That Got the Industry Talking About Risk

The backdrop was an unusual week for the AI industry. Jacob Coxon resigned from his position, and Dario Amodei argued in an essay that AI companies "must slow the pace at which we improve the capabilities of AI models." According to American Banker, this is described as a striking rejection of the race from one of its participants, and the leaders of three other AI producers — Sam Altman (OpenAI), Demis Hassabis (Google DeepMind) and Elon Musk — said they agreed (3937add3).

Against this backdrop, the bank leaders' message came across as a response to the same concern — from those actually deploying the technology at scale.

The Banks' Contained Approach

Moynihan laid out a simple, long-standing policy: Bank of America decided long ago that employees must always check everything AI produces for customers, according to American Banker. No AI agent, therefore, gets to operate autonomously in customer-critical processes.

PNC President Mark Wiedman struck the same tone. "Making sure agents are doing what they're supposed to do and not doing something else is a top priority for us," said Wiedman, president of PNC, which has $616 billion in assets (3937add3). The quote is reproduced here in excerpt; American Banker's article cuts off the quotation itself, so Wiedman's full statement is not available.

The Scale Behind the Assurances — As the Company's Own Numbers

Despite the caution, Bank of America is among the most aggressive AI users in the financial sector. According to Crypto Briefing (b717f1e7), drawing on the company's own disclosures, employees now generate over 400,000 AI prompts daily, the bank has approved more than 300 use cases, reports a 55% reduction in fraud losses and a 20% increase in developer productivity, and its Erica system interacts with 110 internal systems across roughly 700 question types. These figures are the company's own and have not been independently verified.

It is also worth noting that the sources disagree on Bank of America's exact position: American Banker describes it as the country's second-largest bank, while Crypto Briefing calls it the largest US bank by assets (3937add3; b717f1e7).

Why It Matters

Moynihan's and Wiedman's remarks place the banks on a distinct track between the AI industry's warnings and their own technology deployments. They confirm that the risk the industry leaders warn about is real enough to change how even the largest AI users operate: not stopping, but limiting what AI is allowed to do alone. Whether the banks' numbers hold up and the practice works as they describe remains to be independently verified.