AI Risks in Financial Markets and SEC Oversight

With the rising use of synthetic intelligence (AI) in monetary markets, broker-dealers and funding advisers want to concentrate to the dangers posed by AI on corporations’ compliance with federal securities legal guidelines. While machine studying is more and more built-in into in the present day’s monetary ecosystem together with name facilities, compliance techniques, robo-advisers, and algorithmic buying and selling, using AI techniques, and generative AI techniques particularly, are growing quickly.
The mixture of AI’s growth and integration into our monetary markets presents a wide range of distinctive alternatives and challenges for market individuals and regulators. With regulatory curiosity in AI excessive, broker-dealers and funding advisers ought to look at whether or not their present compliance packages handle AI-related dangers adequately.
What is Artificial Intelligence?
With a lot protection of AI in 2023, it could appear unusual to ask: What is AI? AI is an amorphous time period and has been round for the reason that mid-Twentieth century. At its core, AI is a machine/laptop that mimics human intelligence and cognition. Most of us are acquainted with the “skilled system” type of AI that was launched in the Eighties. Expert techniques are packages that clear up issues or reply questions by counting on a system of guidelines coded into this system by consultants. While these techniques emulate human decision-making potential, they’re confined by the boundaries of their programmed information base. Expert techniques are used in making medical diagnoses, laptop gaming, accounting software program, and extra, proving helpful in performing specialised duties however not suited to adaptive drawback fixing.
Machine and deep studying are more moderen types of AI. Machine studying is AI in which the machine solves sure issues by itself, with little or no the enter of human-developed algorithms. This has an apparent benefit over skilled techniques as a result of machine studying just isn’t depending on a programmed information base, and it doesn’t require the laborious growth of algorithms by human programmers. Machine learning-based AI, nonetheless, could require human intervention to be taught and differentiate between knowledge inputs—like how an internet site could immediate you to decide on a motorbike from a collection of images to confirm you might be human. These fashions comply with three main types of studying: supervised, unsupervised, or reinforcement. Under the supervised studying mannequin, human operators practice the mannequin with pre-classified datasets to assist it discover ways to classify knowledge or predict outcomes. Under the unsupervised studying mannequin, the machine will analyze unclassified datasets to find patterns or outcomes with out human intervention. Finally, in reinforcement studying, the machine is educated by way of trial-and-error by in search of a pre-determined motion or consequence.
Deep studying is a extra superior model of machine studying. Deep studying depends on neural networks, comprised of node layers, which emulate neurons sending alerts in a mind. The nodes have at the very least three layers: an enter layer, a number of hidden layers, and an output layer. The nodes can obtain knowledge, analyze it, and formulate an output. In this fashion, deep studying AI can reply extra complicated questions than prior types of AI. Deep studying can digest uncooked knowledge with out the necessity for human intervention to assist it differentiate between knowledge inputs. This offers deep learning-based AI a bonus over machine studying, particularly when working with giant datasets. Users can even mix deep studying with reinforcement studying to investigate giant datasets and optimize outcomes from that giant knowledge. As AI improves, so-called deep reinforcement studying techniques can extra intently emulate the reward-seeking behaviors of the human mind. This guarantees each thrilling and horrifying potentialities for regulated entities and the monetary markets.
Current AI know-how is proscribed to synthetic slim intelligence (ANI), which is designed to carry out a single or slim set of duties. Although we’ve got witnessed super progress in machine and deep studying, AI remains to be restricted to performing simply sure duties with elevated proficiency. Take, for instance, OpenAI’s ChatGPT. ChatGPT is a generative multimodal mannequin restricted to producing pure language responses. And, as attorneys ought to know, ChatGPT can often produce unreliable outcomes.
Researchers like OpenAI are working in the direction of the event of true synthetic normal intelligence, or techniques which are designed to perform any mental activity {that a} human can carry out. Beyond this, AI could finally surpass human intelligence and obtain synthetic superintelligence. The generative AI fashions of in the present day will doubtless appear to be a primitive AIM chatbot in 20 years.
Artificial Intelligence in Financial Markets
Broker-dealers and funding advisers have lengthy used AI instruments in the monetary markets. AI-based purposes have proliferated for makes use of equivalent to operational features, compliance features, administrative features, buyer outreach, or portfolio administration. Chatbots, for instance, present environment friendly and simply accessible help to shoppers, and robo-advisers analyze markets and present funding suggestions to buyers.
Trading fashions are one other standard space the place corporations deploy AI know-how in monetary markets. Quantitative merchants have used algorithmic fashions to determine investments or commerce securities for the reason that Nineteen Seventies. Increasingly, nonetheless, broker-dealers and funding advisers make the most of extra superior machine studying for these functions. Several corporations have launched AI-powered robo-advisers to buyers. For occasion, JPMorgan has publicly introduced its growth of “IndexGPT,” an AI advisor to investigate and choose securities for particular person buyers’ portfolios.
Risks of Artificial Intelligence
Recent advances in AI increase novel dangers for broker-dealers and funding advisers. These dangers embrace, however usually are not restricted to, the next:

Conflicts of Interest. The threat for recognized or new conflicts of curiosity may enhance with using AI, particularly with using robo-advisors or chatbots. AI packages may make funding suggestions to buyers which are extra worthwhile to the agency. Firms could not even perceive how or why AI packages are making these suggestions. With the sophistication of recent AI packages, customers or programmers could not totally perceive the decision-making strategy of the AI. The threat for a conflicted output rises when the willpower of a advice just isn’t readily clear and explainable.
Market Manipulation. AI buying and selling packages powered by machine and deep studying could discover ways to manipulate markets. An AI program designed to attain income with out limitations or defective limitations could discover ways to manipulate markets, for instance, by “spoofing” the market or executing “wash gross sales.” Trading algorithms already brought about the “flash crash” in 2010. It just isn’t far-fetched for AI to begin manipulating markets in the close to future.
Deception. Generative AI is already infamous for getting used to create “deepfakes,” that are real looking photos, movies, or audio primarily based on actual individuals. Bad actors can make the most of these digital forgeries to trigger havoc in the monetary markets by imitating market leaders and delivering pretend information, for instance. Or deepfakes can goal particular individuals by imitating superiors or clients to seemingly authorize actions relating to a buyer’s account.
Fraud. As AI turns into extra built-in into corporations’ funding suggestions and buying and selling choices, there’s a higher threat that unhealthy actors will use confidential buyer buying and selling knowledge for their very own ends. For instance, unhealthy actors may construct a proprietary AI buying and selling program that makes use of buyer buying and selling knowledge to front-run or commerce forward of doubtless market transferring trades.
Data Privacy. AI packages have entry to large swaths of information, together with, probably, private buyer knowledge. AI techniques analyze this private knowledge to be taught and/or make choices or decide outcomes. The assortment and evaluation of enormous swaths of non-public knowledge raises considerations about how that knowledge is used and who has entry to the information.
Discrimination. The chance for unfair remedy, bias, and discrimination is ripe when AI learns from human knowledge. When studying from knowledge saturated with historic biases or racism, AI is prone to choose up its personal biases. This phenomenon is already nicely documented. A 2018 examine, for example, discovered facial recognition packages carried out poorly on individuals of colour.1 From chatbot interactions, hiring choices, or funding suggestions, an AI’s personal bias may cloud its judgment and skew outcomes in illegal or undesirable methods.

Safeguarding AI-Related Risks
Broker-dealers and funding advisers are topic to a wide range of rules implicated by means of AI. Broker-dealers’ and funding advisers’ recommendation and suggestions have to be in the perfect pursuits of their shoppers and they can’t place their very own pursuits forward of buyers’ pursuits.2 Broker-dealers and funding advisers even have overarching obligations to undertake and implement written insurance policies and procedures moderately designed to stop violations of the federal securities legal guidelines.3 Broker-dealers and funding advisers even have an obligation to safeguard buyer data and knowledge.4 Further, federal securities legal guidelines prohibit fraudulent conduct by broker-dealers and funding advisers.
SEC Actions
The U.S. Securities and Exchange Commission (SEC) has already taken observe of the dangers posed by AI. On July 26, 2023, the SEC proposed new guidelines to deal with dangers that AI using predictive knowledge analytics will place a agency’s pursuits forward of buyers’ pursuits.5 Under the SEC’s proposed rule, broker-dealers and funding advisors must consider any use or moderately foreseeable potential use of “coated applied sciences” in any investor interplay, determine conflicts of curiosity the place using the coated know-how would place the agency’s pursuits forward of buyers’, and get rid of or neutralize the impact of these conflicts. The SEC outlined a coated know-how as, “an analytical, technological, or computational perform, algorithm, mannequin, correlation matrix, or related methodology or course of that optimizes for, predicts, guides, forecasts, or directs investment-related behaviors or outcomes.”6 This covers a large swath of AI applied sciences already in use, equivalent to instruments that analyze investor conduct to offer curated analysis experiences or AI that gives tailor-made funding suggestions.
The SEC’s Division of Examinations just isn’t ready for the finalization of this proposed rule. As mentioned in its 2024 Examination Priorities, this SEC Division has already established a specialised crew to deal with rising points and dangers in this space. The 2024 Examination Priorities particularly advise SEC registrants, equivalent to broker-dealers and funding advisers, as follows:
We additionally established specialised groups inside our totally different examination packages, permitting us to raised handle rising points and dangers related to crypto belongings, monetary know-how, equivalent to synthetic intelligence, and cybersecurity, amongst others. Finally, we continued to strengthen our management crew by bringing onboard a variety of key senior and advisory positions and constructing further capability in numerous examination packages to maintain tempo with the quickly growing market ecosystem in step with Congress’ fiscal 12 months 2023 appropriation. (Emphasis added.)7
The Division of Examinations continued:
The Division stays centered on sure companies, together with automated funding instruments, synthetic intelligence, and buying and selling algorithms or platforms, and the dangers related to using rising applied sciences and various sources of information.
SEC Chair Gary Gensler has pursued aggressive examination, enforcement, and regulatory agendas. By all indications, he intends to deal with the intersection of AI and federal securities legal guidelines no otherwise. His speech 9 days earlier than the proposed rule was launched revealed that he’s as well-versed in this space (and in know-how usually) as any chief of a U.S. monetary regulator.8
Looking Forward
Given the present and proposed regulatory framework, it’s critical for broker-dealers and funding advisers to have a agency understanding of the AI instruments they use and then implement applicable insurance policies and procedures for these AI instruments. Firms shouldn’t wait to evaluate their use of AI, together with future use, and put guardrails in place to make sure clients are protected and the corporations fulfill all regulatory expectations.
Firms ought to start by assessing what AI know-how they’re truly utilizing or plan to make use of. After that is full, assessing whether or not such use presents any conflicts of curiosity, potential buyer hurt, or violation of relevant guidelines and rules is really helpful. Firms also needs to contemplate conserving a listing of all of the AI purposes they use, the dangers posed by every AI software, and mitigating controls to deal with every AI-related threat.
Next, corporations ought to implement and periodically overview their written insurance policies and procedures to deal with AI governance and the regulatory dangers posed by AI. Any present insurance policies and procedures could also be equally enhanced to deal with conflicts of curiosity associated to AI, potential buyer hurt, and potential regulatory violations. For instance, corporations could decide to be deliberate and intentional about their use of any new AI techniques, explicitly requiring overview and evaluation of such AI earlier than personnel are permitted to make use of it. Further, supervision by cross-function groups and periodic testing can be useful to grasp how the AI techniques are performing.
Firms also needs to contemplate reviewing their contracts with clients to evaluate whether or not the agency or its distributors have the requisite rights to make use of or share the information. Separately, corporations may wish to consider their contracts with their distributors to see what protections the agency has with respect to the seller’s use of the information shared by the agency in addition to the companies obtained from the seller.
For robo-advisers and AI instruments that present funding suggestions or recommendation, corporations ought to pay specific consideration to the “explainability” of the AI’s suggestions. As AI turns into extra superior, the decision-making course of could subsequently change into extra opaque. As evidenced by the Biden Administration’s current Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, auditability, transparency, and explainability of AI techniques (whether or not or not developed by the consumer) are all important facets of applicable AI growth and use.9  Accordingly, it is very important attempt to perceive the AI decision-making course of and implement applicable guardrails the place wanted. This may embrace periodic testing of robo-advisers, human oversight of suggestions, or limitations on the suggestions. It is equally essential for corporations to make sure that AI know-how just isn’t inserting the agency’s curiosity forward of buyers’ pursuits. Testing and overview will assist be sure that broker-dealers and funding advisers preserve compliance with federal securities legal guidelines and stave off dangers of great examination findings, referrals to the SEC’s Division of Enforcement, pricey litigation, and the corresponding reputational injury to corporations and corporations’ stakeholders.
Firms also needs to place significance on safeguarding and monitoring how AI techniques use buyer knowledge by adopting insurance policies and procedures to make sure that AI instruments don’t misappropriate buyer knowledge for the corporations’ personal use in buying and selling and prohibit who has entry to the client knowledge. Organizations also needs to contemplate updating their written insurance policies and procedures to replicate what buyer knowledge they acquire, how that knowledge is used, how the information is shared, and whether or not applicable buyer consent has been obtained. Finally, no matter AI use, corporations should safeguard towards cybersecurity breaches.
The development of AI in the monetary markets will result in extra consideration from regulators relating to using AI. Accordingly, broker-dealers and funding advisers ought to start to evaluate their use of AI, together with future use, and put in guardrails to make sure that their clients are protected.

1 See Buolamwini, J. and Timnit Gebru, T. “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification,” Proceedings of Machine Learning Research, 2018.

2 See Regulation Best Interest: The Broker-Dealer Standard of Conduct, Exchange Act Release No. 86031, 84 FR 33318 (June 5, 2019); Commission Interpretation Regarding Standard of Conduct for Investment Advisers, Investment Advisers Act Release No. 5248, 84 FR 33669 (June 5, 2019).

3 See Compliance Programs of Investment Companies and Investment Advisers, Release No. IA-2204, 68 FR 74713 (Dec. 24, 2003); FINRA Rule 3110.

4 See Privacy of Consumer Financial Information (Regulation S-P), Exchange Act Release No. 42974, 65 FR 40333 (June 22, 2000).

5 Proposed Rule, Conflicts of Interest Associated with the Use of Predictive Data Analytics by Broker-Dealers and Investment Advisers, Exchange Act Release No. 97990 (July 26, 2023).

7 See “2024 Examination Priorities.” U.S. Securities and Exchange Commission, Division of Examinations, 15 October 2023.

8 See Gensler, G. “Isaac Newton to AI” Remarks earlier than the National Press Club.” U.S. Securities and Exchange Commission, 17 July 2023.

9 Exec. Order No. 14,110, 88 Fed. Reg. 75,191 (Oct. 30, 2023).

Recommended For You