AI, Active Measures and Capital Markets

The Next Evolution of Terrorism

The threat of artificial intelligence is not a distant possibility; it’s a current reality we underestimate.

While Elon Musk, Steve Wozniak, and other well-meaning experts draw attention to the dangers of artificial intelligence (AI), these individuals are focused on artificial general intelligence (AGI). Insiders call it “hard” AI. This is the AI of Terminator, a singular machine program that can complete any task a human can, but more efficiently and creatively.

This technology doesn’t exist yet. And while we should plan diligently and rationally for its arrival (Gladstone AI recently published an informative plan), this focus disguises the threat existing AI algorithms pose today.

Pointillistic or “soft” AI has existed for over a decade. Soft AI is an algorithm that can achieve one objective, e.g., one task, faster, more competently, and in ways previously unthought of by humanity.

Google’s DeepMind’s AlphaGo is an example. AlphaGo is an AI program developed to beat humans in the complex Chinese game of Go. Its creators trained the algorithm on historical human and computer play. In 2016, AlphaGo beat Lee Sedol, who at the time was the second- best player of Go in the world.

While AlphaGo is an incredible advancement in AI technology, it cannot autonomously drive a car, sort laundry by color, or analyze financial statements. AlphaGo is excellent at one task— winning Go games. This lack of adaptability is why the dangers of current AI systems have been pushed to the sidelines of the ongoing AI safety debate.

Building a soft AI model requires access to data, an internet connection, intellectual curiosity, and harvesting open-source AI models abundant across the internet. Online education platforms like Coursera and Udemy offer self-paced courses on computer programming, AI modeling, and mathematics.

Data is the foundation of soft AI models; it is used to train a model’s predictive capabilities. AI's core value proposition is predicting whether a consumer will purchase a good, an election's outcome, or an investment portfolio's returns.

Some experts estimate that building a soft AI model would require a few hundred thousand dollars, a significant discount compared to previous world-changing technologies.

While open-source AI models and democratizing this knowledge will greatly benefit humanity, there is a trade-off for open access.

And only to our peril do we gloss over this significant danger.

Especially when experts focus on the potential adverse effects of an AI program's unintended consequences. As if only moral, enlightened people travel down the road of AI development or that disaster will solely be generated due to an unintentional result of a poorly tested model.

National security professionals, technologists, and foreign policy advisors must ask, what if an AI program is maliciously designed? And who or what organizations could do that?

The first answer is obvious: the nation-state.

One can expect that introducing AI technologies to global militaries will not change why nations fight wars. And nations fight wars to impose their will on an opponent. Once the enemy has complied, the hostilities will cease. One can conclude that maliciously trained AI models employed by governments will be recalled once peace has been declared.

Leading to the second most likely culprit behind the development of a malicious soft AI program – terrorist organizations.

Before diving into the plausibility of a weaponized AI model wielded by terrorists, reflecting on the symmetry of terrorist organizations and AI models is essential. AI is the perfect tool for terrorist organizations to accomplish their political goals and enhance the efficiency of their operations due to their organizational nature.

Terrorist organizations are usually underfunded, constantly under pressure from governments around the globe, operate in austere environments, are demographically homogenous, and focus their efforts on reachable targets, albeit not always their desired ones.

Financially, terrorist organizations must work diligently to procure, obscure, and utilize the funding they do receive. Terrorists are also chronically undermanned compared to their ambitions, do not trust their people, and must be highly disciplined in operational security principles lest they draw the attention of Western intelligence agencies.

This is why the next wave of terrorism will not be hijacking planes or suicide bombings. Nor will terrorists seek to hack into critical infrastructure. These attacks are too time-consuming and require tremendous financial resources and expert human capital to execute. Terrorist organizations have realized that they are not built to sustain these operations. Sending disenfranchised, unemployed youths to their deaths to create enough havoc to be part of the 24- hour news cycle for a few months is an economically unsustainable model.

21st-century terrorists have adapted. ISIS recruiting videos can be mistaken for dystopian Hollywood movie trailers. Hezbollah has a propaganda arm that would make the Stasi blush with envy. Yet, Western nation-states have found ways to blunt these effects with relative success over the last 20 years.

Due to that success, terrorist groups will seek new tactics to attack their enemies that complement their competitive advantages. Terrorists will recede further into the shadows of cyberspace, their ultimate and natural destination. A home that is relatively anonymous, cost-effective, and provides instantaneous global access. This is a logical and inevitable progression.

In the words of the father of guerrilla warfare, Mao Tse-Tung, “The guerrilla must move amongst the people as a fish swims in the sea.”

Cyberspace is the ultimate sea for the deadly predator terrorist fish.

As with all predators, terrorists seek the most significant target that consumes the least energy.

The U.S. financial system is a large target. The government and civilian institutions that make up that system understand this threat, which is why they have hardened their locations in the physical world and the cyber domain. However, they haven’t prepared for an indirect attack that focuses on the behavior of individuals who participate in the market. But economic warfare and market manipulation are not new either.

The first stock exchange started in Amsterdam. In 1602, the first company, the Verenigde Oost- Indische Compagnie (VOC), went public. A natural byproduct of the first modern economy was the realization that one can achieve more by working together and spreading risks. Capital markets exist today primarily to spread capital across job creators to mitigate risk, drive innovation, and create wealth.

Human nature, however, quickly reverted to its anarchic and Hobbesian behavior. The Dutch discovered that certain individuals would manipulate prices through deceit and conspiracy to increase returns. Giving birth to the first regulators to guard against such behavior.

Today, institutions such as the Securities Exchange Commission (SEC) and the Department of Justice (DOJ) actively investigate, prosecute, jail, and fine individuals and institutions that violate securities laws. The government is incentivized to do this to ensure that both retail and institutional investors believe that participants in capital markets operate with the same asymmetry of information. In brute terms, no buyer should have more access to why a seller is willing to sell than any other buyer.

The price manipulation scandal in the Netherlands during the 17th century was quickly handled. The Dutch were able to do this because the information did not have the velocity it does today, the interested parties were limited in size, and its externalities on the broader community were less impactful.

In today’s world, information velocity is instantaneous and uncontrollable, interest in the stock market throughout the nation is high, and companies' ability to raise capital directly affects employment, inflation, and GDP growth.

Because of these factors, the American financial system is a high-value target for malevolent forces. The terrorist threat to the system is unique because terrorists would not aim to increase returns or manipulate prices but to degrade the entire system. They would do this to negatively harm the market's overarching purpose—to improve human flourishing.

Terrorists won’t seek to hack into the New York Stock Exchange, clearing houses, primary dealers, or merchant banks. These institutions have robust cybersecurity teams, protocols, and applications. The expertise and time required to penetrate and damage those systems successfully would be extravagant for a nation-state, let alone a poorly funded terrorist team.

To successfully degrade the financial system, terrorists would aim to distort the foundational functional principle of the market - trust. Terrorists could seek to sow doubt in the independence of the U.S. Federal Reserve, collusion of massive and/or vital firms, price-fixing, or other unethical behavior from investors or government insiders. The news cycle has tremendous effects on the market. A recent example is the recent swing in Lyft’s stock price. A simple typo in earnings guidance pushed Lyft’s stock price by 60% in one day.

Investing is an information game. Investment decisions are made by collecting, synthesizing, validating, and analyzing information. If information becomes unreliable or doubtful, it increases uncertainty, and markets hate uncertainty. Investors don’t deploy capital when uncertain, and businesses can’t invest in projects, hire new people, or pay down liabilities. Poisoning the well of information would have drastic negative impacts on Americans.

The most efficient way to create doubt in the nation’s capital markets is through disinformation. And soft AI is the perfect enhancer to optimize disinformation.

For over a century, intelligence agencies worldwide have utilized disinformation campaigns to achieve specific political outcomes; they call these operations “active measures.”

Some readers may remember the famous “troll factory,” run by a Russian SVR de facto arm and led by the recently departed Yevgeny Prigozhin. It was a classic active measures campaign to influence the outcome of the U.S. 2016 Presidential election. Traditional media sought to inform the public about this operation. However, in doing so, the efforts of this active measures campaign were propagated to a broader audience. The operation was successful because it caused voters to doubt the information they received about both candidates. One of the goals of an active measure campaign is to instill doubt and uncertainty among a target population.

While the Troll Factory incident is a relatively recent example of disinformation, nation-states have utilized and sharpened active measure campaigns in the modern world since the October Revolution of 1917. The hallmark of a successful active measures campaign is to create a believable headline by hiding the lie amongst a mostly truthful statement, targeting groups likely to propagate the lie (referred to as “useful idiots” in the intelligence community), and over a medium that will spread the lie as far as possible.

Historically, the success of active measures campaigns mainly relied on an intelligence officer's knowledge of the local people, culture, and power structures.

A soft AI model removes the need for that knowledge. The model would provide a culturally nuanced and acceptable piece of disinformation that would not cause an average individual to doubt its veracity. Terrorists can web-scrape from Instagram or hook into the API of X to gather the proper training data to conduct their unique “sentiment analysis.” Useful idiots, motivated by political views, will push information that confirms their own biases throughout the most significant town squares in the history of the world. Algorithms on social media platforms will help spread these headlines across the internet with such speed that proves Virgil’s thoughts on the subject centuries ago, “Rumor, than whom no other evil is faster.”

Compounding this logical evolutionary development for terrorist campaigns is the political climate in America. Traditional news media viewership and trust have dropped significantly in the last ten years. According to a Gallup Poll, 38% of Americans have no trust that mass media “fully, accurately, and fairly” report the news, while only 28% have a “fair amount” of trust that they do. Further, U.S. adults under 30 are as likely to trust information on social media as they are from national news outlets. While some may find the dying nature of mass media to their delight, traditional news media had processes and controls to ensure sources were credible. They produced standards for due diligence, valued truth over speed, and staked their reputations on their journalistic ethics.

The loss of these institutions opens the nation to manipulation and attack.

Imagine this scenario. You are a portfolio manager for CalPERS. (The largest pension fund in the United States, with over $480 billion AUM.) Your Bloomberg News App is vibrating incessantly. You pick up the phone to see the headline, “Whistleblower claims U.S. Treasury Department

Manipulating 10-year Bond Rates”. Your handler at Morgan Stanley is calling you. He tells you this is false; he has no idea what is happening, but he is meeting with other primary dealers at the NY Fed within a half-hour. Meanwhile, your positions are getting crushed.

A few hours later, it turns out that Bloomberg ran with the story because a well-respected and highly followed finance figure on X reposted the headline. But, to keep up with the speed of news on social media, Bloomberg propagated the message without conducting proper due diligence. Markets recover, but you still wonder if the Treasury Department is manipulating the sealed bid process due to the debt calamity.

Two weeks later, you receive a notification from X that “Congressional Aides Charged With Insider Trading” is trending. A few hours later, CNN runs a clip claiming, “Money Manager Vanguard Under Investigation by SEC by falsifying NAVs”. (One can get the point.)

The terrorists behind this disinformation campaign do not understand how the Treasury Department sells bonds, the yield curve's importance, or the essential role of bonds in government funding.

They fed a corpus of text from Barrons, Financial Times, The Wall Street Journal, and other finance literature into an AI model that is correlated to movements in the S&P 500. Then, they had the model predict news headlines affecting the market.

Or they could skip that work, pay $49 a month to Longshot.ai, and have them generate the headline and body of work based on a few keywords. (Watch the how-to video, it’s impressive.)

Terrorists have seen the effectiveness of political disinformation in the United States. It would be an error to underestimate the intelligence and self-awareness of terrorist organizations. Indeed, supporters of ISIS recently released a guide to protect one’s anonymity when accessing ChatGPT.

It is only a matter of time before terrorist organizations realize this moment in American history.

Americans are polarized, gullible, disenfranchised, and angry. The nation no longer has gatekeepers for the validity of news or information. Active measures campaigns have been successful in an adjacent vein through politics. The skills and ingredients to create a soft AI model are accessible, relatively inexpensive, fortify terrorists' organizational structures, and enhance their natural operating habitat.

Capital markets are prepared for cyber and physical attacks but not for a systemic, calculated, persistent active measures campaign from an organization that aims to degrade the system, not manipulate it. By disrupting the information cycle, terrorists would instill doubt, uncertainty, and poison into the capital markets system.

Osama Bin Laden’s famous letter explaining why Al-Qaeda attacked the United States on September 11^th displays a lack of understanding of how America funds itself global position. He writes, “The American people are the ones who pay the taxes which fund the planes that bomb us in Afghanistan, the tanks that strike and destroy our homes…”

Osama Bin Laden did not understand that taxes are simply the top layer of a complicated system. The true power lies within the American financial system. Capital markets allow for servicing of the U.S. national debt, corporate investment, employment, and economic growth.

We cannot hope for that misunderstanding in the future. The terrorist goal to harm America and its way of life still exists. The best way to accomplish that mission has emerged.

Access to data, the ability to self-study soft AI modeling, the current domestic political mood, geopolitical instability, a vacuum of trustworthy information, and the nature of terrorist organizations will increase active measure campaigns from non-nation state actors to harm the financial system.

To paraphrase Stuart Russel, the goal of an artificial intelligence program is to take inputs from its perceptions and achieve the objective(s) set by its creator. A.I. is not inherently good or evil. It’s a tool.

But the danger of that tool isn’t coming. It’s already here.

This article was written in March 2024.