A New Frontier: The SEC Addresses Artificial Intelligence (and a Whole Lot More)

Executive Summary

• Newly proposed rules (the Predictive Analytics Rules¹) will, if adopted as proposed, require broker-dealers and investment advisers (firms) to: (i) identify conflicts of interests when using certain technology in interactions with “investors”; and (ii) adopt policies and procedures that eliminate or neutralize (rather than disclose or mitigate) those conflicts of interests.
• The Predictive Analytics Rules would cover a broad range of technology that includes not just “artificial intelligence,” but also any other analytical, technological, or computational function, algorithm, model, correlation matrix, or similar method or process that optimizes for, predicts, guides, forecasts, or directs investment-related behaviors (Covered Technology).
• The proposal would apply in situations where firms use Covered Technology in connection with the firm’s engagement or communication with an “investor,” including by exercising discretion with respect to an investor’s account, providing information to an investor, or soliciting an investor.  This includes communications that take place in person, on a website, via smartphones, through computer applications, or by email, among other modes of communication (including advertisements).
• Because of the manner in which the term “investor” is defined, the Predictive Analytics Rules apply differently to broker-dealers and investment advisers. Specifically, the Predictive Analytics Rules will apply with respect to Covered Technology used in interactions with all clients of an investment adviser, including institutional investors, as well as investors in registered funds and certain private funds. For brokers-dealers, the Predictive Analytics Rules, however, would only apply with respect to interactions with retail customers.

Introduction

The Predictive Analytics Rules proposed by the Securities and Exchange Commission (SEC) are intended to address conflicts of interests associated with the use of certain technology in connection with broker-dealer and investment adviser interactions with investors.² If adopted as proposed, the Predictive Analytics Rules will require firms to: (i) identify conflicts of interests when using certain technology in interactions with investors, and (ii) adopt policies and procedures that eliminate or neutralize (rather than disclose or mitigate) those conflicts of interests.

In connection with the proposal, SEC Chair Gary Gensler stated that predictive data analytics technologies provide an increasing ability to make predictions about individuals, which increases the risk that firms will put their interests ahead of clients or customers without an effective way of addressing the related conflicts.³ However, the Predictive Analytics Rules impose very prescriptive requirements over a very broad array of technologies and conduct that could have adverse impacts on the adoption of new technologies.

Comments on the Predictive Analytics Rules are due 10 October 2023.

Overview of Predictive Analytics Rules 

The Predictive Analytics Rules would establish new requirements for firms making use of certain technology when interacting with “investors” (as defined in the proposal):

Identifying Covered Technology and Investor Interactions

The Predictive Analytics Rules would require firms to identify certain technology (Covered Technology) used in investor interactions (Investor Interactions). The Proposing Release defines Covered Technology as “an analytical, technological, or computational function, algorithm, model, correlation matrix, or similar method or process that optimizes for, predicts, guides, forecasts, or directs investment-related behaviors or outcomes.”⁴ An Investor Interaction is defined as “engaging or communicating with an investor, including by exercising discretion with respect to an investor’s account; providing information to an investor; or soliciting an investor; except that the term does not apply to interactions solely for purposes of meeting legal or regulatory obligations or providing clerical, ministerial, or general administrative support.”⁵ As discussed in greater detail below, because the definition of "Covered Technology" includes a broad range of technology used to “guide” (an undefined and vague term under the proposal) investment-related behavior and Investor Interactions include, among other interactions, merely providing information, the scope of the proposal could potentially be extremely broad in the absence of further clarifying guidance. 

For purposes of the proposal, the term “investor” is defined differently for advisers and brokers. For advisers, it is defined as “any prospective or current client of an investment adviser or any prospective or current investor in a pooled investment vehicle (as defined in Rule 206(4)-8 under the Investment Advisers Act of 1940 (the “Advisers Act”)) advised by the investment adviser.”⁶ The scope, while similar, is broader than that of Rule 206(4)-1 under the Advisers Act, as it would apply to interactions with investors not just in private funds relying on Section 3(c)(1) or Section 3(c)(7) of the Investment Company Act of 1940, but also to registered funds.⁷ For brokers, the term “investor” is defined in a manner that is consistent with the definition of a “retail customer” under Regulation Best Interest⁸ as “a natural person, or the legal representative of such natural person, who seeks to receive or receives services primarily for personal, family or household purposes.”⁹ 

Identifying Conflicts of Interest

Once Covered Technology used in Investor Interactions has been identified, the Predictive Analytics Rules would require a firm to evaluate “any use or reasonably foreseeable potential use” of a Covered Technology that may give rise to a conflict of interest “associated with that use or potential use” of a Covered Technology.¹⁰ As discussed in greater detail below, this requirement could create significant hurdles for the use of certain technologies, where the methodology used by the technology to generate an output cannot reasonably be evaluated.

Triaging and Eliminating Conflicts of Interest

Once these uses of Covered Technology in Investor Interactions have been identified, a firm would need to: (a) determine if that conflict places the interest of the firm ahead of the investor, and (b) eliminate or neutralize the conflict of interest that places the firm’s interest ahead of the investor’s interest.¹¹ Critically, this requirement does not permit firms to disclose or mitigate conflicts in the manner clearly articulated by the SEC in prior guidance related to advisers’ fiduciary duty and broker-dealers’ standards of conduct under Regulation Best Interest.

Written Policies and Procedures

The Predictive Analytics Rules would also require the adoption of written policies and procedures reasonably designed to prevent violations of the Predictive Analytics Rules. 

All firms’ policies and procedures would need to include written descriptions of: (a) the process used to evaluate any use or reasonably foreseeable potential use of Covered Technology in any Investor Interaction, (b) the process used to determine whether any identified conflict of interest results in an Investor Interaction that places the interest of the firm ahead of the interests of the investor, and (c) the process used to determine how to eliminate, or neutralize the effect of, any such conflicts. Certain firms that use complex technology may also need to include additional elements, including: (a) compliance review and monitoring systems and controls, (b) procedures that designate responsibility to appropriate personnel for supervision of functions and persons, (c) processes to escalate identified instances of noncompliance to appropriate personnel for remediation, and (d) training of relevant personnel on the policies and procedures, as well as the forms of Covered Technology used by the firm.¹²  Additionally, the policies and procedures of all firms would be required to include a provision for reviewing, no less frequently than annually, the adequacy of the policies and procedures and their effectiveness.¹³ 

Recordkeeping Amendments

The Predictive Analytics Rules also set forth requirements for firms to maintain records related to their use of Covered Technology.¹⁴ The proposed rules would mandate that a firm maintain written documentation of: (a) the firm’s evaluation of Covered Technology to determine what uses would give rise to a conflict of interest, (b) the firm’s determination whether any use of Covered Technology places the interests of the firm ahead of the interests of the investor, and (c) how that conflict of interest has been eliminated or neutralized. 

In addition to the written documentation discussed above, firms will be required to maintain a record of the firm’s: (a) annual review of the adequacy of their policies, (b) any disclosures provided to investors regarding the firm’s use of Covered Technology, and (c) records of each instance in which a Covered Technology was disabled and why it was disabled.

Key Takeaways and Open Questions for Advisers and Brokers Regarding the Predictive Analytics Rules

While the SEC has taken the position that the Predictive Analytics Rules are narrowly tailored to prevent firms from unfairly influencing the investment activity of their investors, the proposed regulations have already given rise to criticism not only from the finance industry but also from two SEC commissioners at the open meeting where they were proposed. 

The Definition of “Covered Technology” Will Capture a Broad Array of Technology 

The Predictive Analytics Rules define Covered Technology extremely broadly. If adopted as proposed, the Predictive Analytics Rules would require firms to spend significant time and resources evaluating nearly all of the “technology” used to determine whether the software, devices, or other technologies fall under the rubric of Covered Technology and implicate these requirements. 

As part of these evaluations, firm compliance and legal personnel would need to engage with their technology and business colleagues to evaluate whether a technology is Covered Technology and then determine whether it is being used in connection with an Investor Interaction, even if such interaction is not a “direct” interaction with investors. This process would even include communicating with third-party service providers that supply a firm with Covered Technology. 

In sum, the Predictive Analytics Rules would automatically impose a lengthy and involved evaluation process regarding new and existing technologies that would take up significant firm time and resources and would furthermore be ongoing in nature.

The Definition of “Investor Interactions” Will Capture Most Interactions

The Predictive Analytics Rules also define Investor Interactions broadly. This definition of "Investor Interactions" would capture a firm’s correspondence with or conveyance of information to investors regardless of form. This includes communications that take place in person, on a website, via smartphones, through computer applications, or by email, among other modes of communication (including advertisements). 

The limitation of the regulations to Investor Interactions is supposed to recognize that technology used in investor-facing contexts is more liable to unfairly manipulate an investor’s investment decision; however, the definition is so broad (particularly when coupled with the broad definition of Covered Technology) that it could be read to cover all of a firm’s technology except for “purely ministerial or back office” processes.¹⁵ 

The Potential Breadth of Conflicts of Interest to Be Eliminated or Neutralized Could Require Changes in Existing Business Models

The Proposing Release clarifies what qualifies as a conflict of interest, and the rules around what qualifies as a conflict of interest could also have an overly broad application in practice. 

For example, many investment advisers create models of a company’s financial statements to evaluate whether to advise their clients to invest in that company. If such a financial model considers metrics that are favorable to the adviser (such as a potential performance-based fee), the use of this model would include a conflict of interest. The firm would have to evaluate the specific circumstances of the use of the model and determine whether the conflict of interest results in the adviser’s interest being placed ahead of an investor’s interest. Although, a conflict arising from this model would not be prohibited under the Predictive Analytics Rules if it can be neutralized or the outcome is equally favorable to the investor regardless of whether the factor (like a performance fee) is considered, making the determination would likely require reviewing a broad universe of alternative investment opportunities.¹⁶ 

If a model is designed to completely screen out an investment if it would not result in a fee for the adviser, then that would be an example of the adviser’s interest being placed ahead of an investor’s interest. This is because the investor is being deprived of an investment opportunity as a result of the adviser’s use of the model.¹⁷ Many firms, however, currently make use of models that require that a certain percentage of the model portfolio include affiliated products. Under the proposal, in such instances firms would likely be required to offset fees for all clients in a manner similarly required for retirement accounts under the Employee Retirement Income Security Act of 1974 (ERISA).¹⁸

The same conflicts would arise for a broker-dealer using an algorithm to recommend investment ideas to investors. Brokers would be required to analyze algorithms used in connection with such recommendations and determine that they do not give extra prominence to investment ideas that will give rise to revenue for the broker or otherwise eliminate or neutralize the conflict.

How Do Firms Eliminate or Neutralize Prohibited Conflicts of Interest?

The test for whether a firm has successfully eliminated or neutralized the effect of a conflict of interest is whether the interaction no longer places the interests of the firm ahead of the interest of investors.¹⁹

Imposing this new standard is a significant deviation from existing practice. When conflicts of interest are present in a relationship, firms have historically been permitted or required to provide full and fair disclosure of the conflict and obtain informed consent from the client. This practice has been reflected in both the Commission Interpretation Regarding Standard of Conduct for Investment Advisers²⁰ and Regulation Best Interest,²¹ each of which was adopted in 2019. In those instances, elimination of the conflict is only required when it cannot be addressed through disclosure or mitigated. 

Nevertheless, in the Proposing Release, the SEC implies that disclosure of a conflict of interest will not be effective for Investor Interactions with Covered Technology. Specifically, the Proposing Release notes that the size of the datasets, the complexity of the algorithms on which the Covered Technology is based, and the ability of the technology to learn investor preferences or behavior would result in disclosures that are highly technical and could be difficult for investors to understand.²² Accordingly, in the SEC’s view, investors are not capable of understanding the disclosure and provide informed consent, so firms may no longer disclose away conflicts in these cases. When applied in the context of other regulatory obligations, such a requirement may alter how firms conduct their business in areas that are otherwise outside the scope of the Predictive Analytics Rules. Firms may also choose to question the SEC’s statutory authority to mandate such sweeping changes on this portion of the Predictive Analytics Rules during the comment period.

Moreover, the Predictive Analytics Rules draw seemingly arbitrary distinctions regarding how disclosing conflicts of interest can render these conflicts innocuous. For example, under the current regime, firms may make use of a model in Investor Interactions that recommends the use of affiliated funds as an investment option. In such instances, so long as the firm has provided full and fair disclosure of the fees and other conflicts associated with the use of affiliated funds as an investment option and received the informed consent of the investor, the model may be used, and the conflicts of interest are not deemed to be a violation of the firms fiduciary duty in the case of an investment adviser or the Regulation Best Interest obligation in the case of a broker-dealer. By contrast, if a firm makes use of Covered Technology that does the same thing in a way that is subject to the Predictive Analytics Rules, then the investment adviser or broker-dealer would be required to fully neutralize or eliminate the effects of the conflict, which may require the firm to offset the fees associated with the affiliated funds. 

If these regulations are adopted, legal and compliance personnel at firms using Covered Technology will need to carefully consider whether they can eliminate or neutralize a conflict of interest where the firm’s interest is placed above an investor’s interest. If firms cannot do so, they may need to cease use of the Covered Technology. The Predictive Analytics Rules may also cause firms to forego making use of technological advancements that could be beneficial to investors and reduce fees, given the complexity required to comply with the Predictive Analytics Rules.

The Predictive Analytics Rules Will Prohibit or Require Substantial Changes to Some Types of Covered Technology 

The financial services industry has rapidly adopted highly complex computational technology. Some more complex Covered Technology may lack explainability as to how the technology functions in practice, and how it reaches its conclusions (i.e., “black box” algorithms).  As a practical matter, firms that use such Covered Technology may be unable to identify all conflicts of interest associated with the use of such Covered Technology, unless such firms can modify the technology to embed explainability features into their models or adopt back-end controls. 

While, in certain cases, developers may embed explanatory features into their models or adopt back-end controls, in some cases this is highly impractical or even impossible. Accordingly, firms may need to completely prohibit the use of this technology should the new rules be adopted. Legal and compliance professionals will need to be prepared to have detailed conversations with senior leaders at firms as to why certain technologies are inappropriate under the Predictive Analytics Rules, if adopted as proposed.

Not All Covered Technology Will Need the Same Level of Review

Generally, under the Predictive Analytics Rules, a firm can adopt different approaches for different Covered Technology, depending on the nature of the Covered Technology, how it is used, and the firm’s plans for future use. For example, a firm that uses a “simple” Covered Technology, such as basic financial models or simple investment algorithms, could take simple steps to evaluate the technology and identify any conflicts of interest. These steps could include reviewing the Covered Technology to confirm whether it weighs outcomes based on factors that are favorable for the firm, such as the revenue generated by a particular course of action. 

For more “complex” Covered Technology, such as those that use artificial intelligence or machine learning, the firm may need to take more extensive steps to evaluate the technology and identify any conflicts of interest. As noted above, merely disclosing the conflict may not be sufficient. These steps may include conducting a risk assessment, reviewing the technology’s code (if possible), and testing the technology’s performance. 

Firms heavily dependent on Covered Technology will be more burdened by the compliance costs associated with the Predictive Analytics Rules than peer firms that utilize less Covered Technology. This could place small firms, or firms that use Covered Technology to minimize costs, at a competitive disadvantage. Legal and compliance personnel will need to work closely with their business colleagues to evaluate the ongoing compliance costs associated with bringing a new product (or modified product) to market. Firms may determine that some tasks are better suited for a human, rather than Covered Technology, in order to limit compliance costs associated with the Predictive Analytics Rules.

Difference in Scope Between Broker-Dealers and Investment Advisers

The Predictive Analytics Rules apply differently to broker-dealers and investment advisers. Under the Predictive Analytics Rules, the prohibitions related to Covered Technology apply to all clients of an investment adviser, including institutional investors. In contrast, a broker-dealer would not be required under the Predictive Analytics Rules to take the affirmative steps to neutralize or eliminate bias in Covered Technology if the application is facing institutional investors, though they would be required to do so if the application faces a retail customer. 

Whether an institutional client of an investment adviser needs more “protection” than an institutional client of a broker-dealer is an area that will likely be commented on by industry participants. By eliminating the ability to cure a conflict of interest via full and fair disclosure, institutional clients of investment advisers may receive fewer benefits from Covered Technology than their institutional customers investing with a broker-dealer. We expect that many firms will advocate for a more flexible standard, including an exemption from the Predictive Analytics Rules for institutional investors, during the comment period.²³

Next Steps

Given the broad scope of the Predictive Analytics Rules, firms should carefully assess this proposal against their own business practices and operations. Interested investment advisers, broker-dealers, technology providers, and other industry participants may wish to consider submitting comments discussing their views regarding the Predictive Analytics Rules, including the matters described above. In particular, the SEC has solicited comments concerning scalability, which they believe could rapidly exacerbate the magnitude and effect of conflicts, such that it would make full and fair disclosure unachievable. 

In addition, firms should expect the SEC and other financial service regulators to use existing antifraud mechanisms and fiduciary standards to confirm firms are not using Covered Technology to put the interests of their firm ahead of the interests of their investors. Firms should review and evaluate their existing disclosures as they pertain to their use of Covered Technology and whether the disclosure appropriately identifies and explains any conflict of interest or perceived conflict of interest. As part of this process, lawyers and compliance personnel may also want to deepen their understanding of how their firm is using Covered Technology and how it interacts with investors.