Cinch Up! AI Enforcement Starts With Washing Charges
In March 2024, the SEC announced that it settled two cases against investment advisers Delphia (USA) Inc. and Global Predictions Inc. for making false and misleading statements about their purported use of artificial intelligence, when they were not using the AI as marketed. The firms agreed to settle the SEC’s charges and pay $400,000 in civil penalties.
For months, SEC leadership has made repeated warnings about "AI washing," i.e., falsely claiming the use of AI or machine learning models or misrepresenting their application. The March enforcement actions are the first in which the SEC has charged companies with violating federal securities laws in connection with AI disclosures. We can expect many more.
The use of AI-driven tools by investment companies and professionals is not new. For more than a decade, algorithmic trading, high-frequency trading, AI-managed portfolios, and other AI-powered systems have started to eclipse traditional investment platforms. Recently developed AI, however, increases the speed around the track, because it can be trained on multiple types of data to understand how and why markets behave as they do, process enormous amounts of information in mere seconds, and predict market trends with sometimes impressive accuracy. On the other hand, AI is subject to error and bias from, among other things, the data inputs selected — sometimes called “GIGO” for “garbage in, garbage out.”
SEC enforcement actions for misleading disclosures are not the only potential pitfall of AI use. Other potential AI abuses threatening to disrupt financial markets include:
- Conflicts of Interest. AI use by broker-dealers and investment advisers may pose increased conflicts of interest if used in a manner that results in the firm placing its own interests above its investors’ interests. For example, one type of conflict may arise where an investment professional employs a proprietary AI algorithm to make investment decisions, which the investment professional has a financial interest in keeping confidential. Additionally, AI-powered tools are often a “black box” to the professionals that employ them, as such professionals are often unaware of how AI reached certain conclusions, thus resulting in a possible breach of fiduciary duty. If the algorithm is tainted, for example, by the use of corrupt or biased (GIGO) data, then the capacity to quickly scale information could become problematic as the transmission of any resulting recommendations could spread rapidly to many investors, causing unexpected market impact. The risk, among others, of such “algorithmic error amplification” prompted the SEC to propose a sweeping AI conflict of interest rule in July 2023, which has not yet been finalized. See “SEC Proposal Balances AI-Like Technology Use With Investor Best Interests – Has the SEC Picked a Winner?” Expect Focus – Life, Annuity, and Retirement Solutions (September 2023).
- Market Manipulation. Through various methods (i.e., deep learning or reinforcement learning), AI algorithms may learn how to manipulate and destabilize securities markets and avoid detection, when programmed with malicious intent. For example, a trader could deploy an AI-powered trading “bot” to create false market demand or supply, thus artificially inflating or deflating prices to the trader’s advantage, and then avoid detection by mirroring non-threatening algorithmic trading systems.
- Deception. For quite some time, bad actors have used AI to clone voices, alter images, and even create fake videos to spread false or misleading information, also known as “deepfake” AI. Fraudsters can use these digital forgeries to impersonate a financial professional’s customer or to disrupt financial markets by imitating powerful individuals in a multitude of schemes. For example, in “pump-and-dump” schemes, deepfake imagery or audio can drive up stock prices while fraudsters unload their shares before the ruse is uncovered. In 2023, a deepfake image of a billowing black cloud above the Pentagon caused the Dow Jones Industrial Average to fall approximately 80 points in four minutes, which illustrates how powerfully even unsophisticated spoofs can impact financial markets.
Public companies, investment advisers, and other investment professionals that use AI should prepare for enhanced SEC scrutiny, even in the absence of specific SEC AI rules. The track ahead is long, but the SEC has left the starting gate.
