Online Fakery and Digital Bank Runs Are a Scary Mix

There’s no regulatory toolkit to confront deepfakes and misinformation.

Potential safeguards could include more early warning systems or circuit breakers  of some kind to pause withdrawals. (Getty Images/iStockphoto)

 Hackers crack the Twitter account of the Associated Press. They post a tweet: Two explosions at the White House, the president is injured. The S&P 500 instantly drops 1%. Sound like fiction? This happened almost exactly a decade ago. Thankfully, the market recovered quickly after Barack Obama's press secretary confirmed no explosions and no injury.

The world of online fakery has moved on a lot since then -- and not for the better. Advances in artificial intelligence have made more sophisticated chatbots and bogus photos and videos easier to produce and propagate. Meanwhile, bankers are quickly realizing that people can move their money faster than ever before through a few quick taps on their smartphones. Everyone, everywhere needs to be as vigilant as possible to the potential for catastrophic cons in finance right now. Markets are already skittish and depositor psychology is unsettled.

Silicon Valley Bank lost $42 billion, or more than half its demand deposits, within hours when its run began. Regulators and executives have noted social media's role. Citigroup Inc. Chief Executive Officer Jane Fraser recently called the combination of mobile-banking apps and social media a game-changer for bank stability when discussing SVB.

“There were a couple of tweets and then this thing went down much faster than has happened in history,” she said at an Economic Club of Washington event last week.

SVB was arguably a special case: It had a highly concentrated deposit base over which powerful individuals such as tech mogul Peter Thiel had real influence. The run might have happened almost as quickly without social media.

Credit Suisse Group AG, the now defunct Swiss bank, was perhaps also a special case: Years of scandals and management failings had undermined trust and rattled staff and clients. Still last October, it had more than enough capital, no portfolio of bad assets and more than double the cash and liquid assets it needed to meet a stressed level of deposit outflows.

And yet, early that month when executives were in the midst of a strategy review, one weekend tweet claiming that an unnamed European bank was “on the brink” blew up a social media storm and Credit Suisse's customers pulled a shocking amount of assets, nearly $90 billion, in a few weeks. The bank was able to meet those withdrawals without having to sell any assets at a loss. But six months later, a few more bad headlines in a panicky market were enough to kill it.

Federal Reserve Chairman Jerome Powell talked about the speed of the SVB bank run at last week's monetary policy press conference, saying “it's very different to what we've seen in the past.” There might need to be regulatory changes to keep up with what's happening in the world, he added. Potential safeguards could include more early warning systems or circuit breakers  of some kind to pause withdrawals.

If speed is a danger, misinformation is a potentially deadly toxin. Fake images and videos are becoming ever more difficult to distinguish from reality. Tools like Stable Diffusion and Midjourney started becoming popular late last year and have become rapidly more sophisticated.

Last week, fake images of Donald Trump getting arrested and Emmanuel Macron running between riot police spread across Twitter and Facebook. Both were easily debunked. But another fake image of Pope Francis wearing a designer puffer jacket was viewed more than 4 million times and many say they were taken in. Unlike the photos of the 76-year-old Trump sprinting, the hipster Pope is just plausible enough to be believable. Culture writer Ryan Broderick called it the “first real mass-level AI misinformation case.”

Now, imagine a deepfake video of a widely recognized central banker or bank executive, such as the Fed's Powell or JPMorgan Chase & Co.'s Jamie Dimon, saying something plausible but troubling about a major lender. A digital bank run is a whole lot harder to stop than to start.

The usual suspects for fakery have had their interests piqued. An analysis of internet forums known for promoting conspiracy theories like QAnon shows a number of visitors experimenting with image-generating tools. Some have swapped tips on how the technology can be used and amplified. A serious, coordinated effort to disrupt markets is bound to come. 

Last year, Facebook and Twitter showed marked improvement in stopping bots from Iran and China from sowing disinformation on their platforms ahead of the midterm elections. Since then, Elon Musk has gutted much of the team at Twitter that was stopping such manipulation.

And on top of sophisticated and free image generators, large language model technology – which powers tools like ChatGPT – is also more accessible to propagandists, scammers, and other bad actors. OpenAI's own System Card for GPT-4, which analyzes all the safety challenges of the technology underpinning ChatGPT, admits that it “has the potential to cast doubt on the whole information environment, threatening our ability to distinguish fact from fiction.”

Banks are stronger since the 2008 financial crisis and the largest are much more able to meet a sudden jump in withdrawals. But that didn't save  Credit Suisse. Ultimately, a certain level of trust is critical to banks' survival and doubt is like kryptonite. Keep your wits about you.