Generative AI is already manipulating financial markets

Yesterday a fake image of an explosion at the Pentagon was circulated on a verified Twitter account and was reposted by several other accounts before being rebuffed by local law enforcement and others who noted that the image was clearly generated. However it spooked the markets and was promoted at a timely manner given what was kicking off on the Ukrainian-Russian border.

Market manipulation from Generative AI and useful idiots like verified accounts on Twitter that align or mimic trusted sources will become increasingly prevalent. If we can easily fall for a generated image the markets will definitely be susceptible when AI starts employing more complicated manipulation efforts.

I can see 5 immediate methods Generative AI will use to manipulate a market with the instigators seeking to make profit from the resulting price movement:

Disaster Reporting

Generating a disaster is now a doddle with the ability to create a collection of images from different angles with the same style of the same event. This can be reinforced with generated reporting and witness statements. Amplification is then performed on verified social media accounts at the optimal time. All of the required assets will be completely orchestrated, generated, posted and amplified by the AI and will be timed to respond to real world events to increase plausibility and escalation.

The reporting of the Pentagon incident briefly wiped 30 points off the S&P 500 or about $500 Billion in market cap.

Almost all Generative AI market manipulation will create a downward price movement or a relatively small short term sell off. From a trading perspective its a pretty low risk game - set up the trade to trigger a SELL order from the target instrument. A lot of money can be made and is relatively hard to detect because at any given time it is reasonable to have a perfectly innocent trade that would align to the market manipulation.

Scandal Manufacturing

This will be all about targeting specific executives within publicly traded companies with personal or professional scandal or accident, think Elon being in a car crash or Bob Iger being tied up in the Jeffery Epstein scandal. Now, it’s not like these fake stories haven’t been around for years, however it will become harder to pick the fakes as images, copy, reporting becomes hard to distinguish from typical coverage. The market manipulator only needs minutes of overreaction to clean up and walk.

Analysis Revisionism

We can now consume and analyse vast swathes of fundamental and technical data about an instrument (a stock equity, for example) in seconds and create a detailed analysis of the company fundamentals such as sales figures, product strategy or industry factors or indeed the technical analysis on the instrument price. It is relatively easy to modify the context that the model uses to generate the analysis in order to generate a modified view from the baseline truth, either a more positive or more negative outlook. The depth and sophistication of the analysis will fool most readers and is especially dangerous if disseminated into the financial content ecosystem.

Malfunction Amplification

Recalls, service down, injury, fire and failure are all issues that are normally reported on in due course and mostly accurate to the original press release, and can all have significant impact on price - think of people dying on Peloton treadmills for impact of share price. An absolute gold mine for generative AI to co-ordinate an attack on a specific product and company, lets say something simple like an iPhone exploding on a plane with images and tweets and press release from Apple that are all created and orchestrated to piggy back an actual incident on a flight where multiple people report smelling and seeing smoke in the cabin.

Community Manipulation

You can now build and run an entire financial trading community from an AI, fake moderation, fake analysis, fake technical analysis - this is a build on an age old process of pump and dump, except now AI is being used to create the buzz and shape the narrative and importantly create detailed analysis capable of shifting the price of smaller cap securities. Manipulation at scale without the boiler rooms.

Trust & Regulation

I anticipate 90% of content in finance will be generated by AI in the next few years - the story might be real but the copy that we read will be created and therefore more easily ‘optimised’ to illicit a certain reaction from the reader and from the market. Trusted sources and the suppliers such as banks, brokers, exchanges etc. will be more important when verifying a market story, however, much of the advantage may be lost in the time that this will take. Regulation will be key to managing this phenomenon before market integrity begins to erode.

The march to discredit official news outlets and drive news to the social platforms including the Elon Musk ‘truth’ agenda which relies on community notes will prove ineffective against ever more convincing market stories designed to mislead and manipulate the markets for personal gain.

What can we do?

Not much. It’s going to be harder not easier to identify deep fake stories. However I think one of the keys to breaking down plausibility of these stories is to visually identify AI generated images. I think there is multiple benefits to cracking this nut but it would certainly go someway to limiting the success of many of the above methods as we buy with our eyes so much. A digital and physical watermark underpinned by an NFT for example and rendered at the browser or UI layer might restrict the ability to build momentum with emotive imagery.

Regulating this will be difficult but if we are being honest any AI legislation now feels a little like the horse has left the solar system let alone bolted out the gate.