Regulators in the United States and abroad are showing increasing interest in pursuing enforcement actions against companies that deploy artificial intelligence, machine learning, or algorithmic-based applications (“AI”) in a way that the regulators perceive as harmful to the public. These regulators expect transparent and comprehensive disclosures by companies regarding AI that can negatively affect clients or customers. The United States Securities and Exchange Commission’s (“SEC”) recent enforcement action against BlueCrest Capital Management (“BlueCrest”) highlights the risks of not disclosing the use of an algorithmic trading tool. Although the SEC had additional concerns, our focus here will be BlueCrest’s use of algorithms.

Background

On December 8, 2020, the SEC entered into an offer of settlement with BlueCrest for $170 million related to conduct occurring between 2011 and 2015. The SEC order found that BlueCrest, based in the UK, violated the negligence-based antifraud provisions of the Securities Act of 1933 and the Investment Advisers Act of 1940 by failing to adequately disclose an algorithm it developed as a substitute for live traders, in addition to making material misstatements and misleading omissions regarding its use of the algorithm in its Form ADV, offering documents, responses to investor due diligence questionnaires, and discussions with due diligence consultants.

We will discuss facts here as alleged by the SEC in the settled Order, noting that BlueCrest has not admitted or denied these findings. In 2011, BlueCrest created a proprietary fund, BSMA Limited (“BSMA”), to trade capital invested by BlueCrest Executive Committee members. Significantly, BSMA employed a similar investment strategy to its BlueCrest Capital International Fund (“BCI”). Shortly after creating BSMA, BlueCrest began transferring top-performing traders from BCI to BSMA. Instead of replacing the traders in BCI, BlueCrest partially reallocated their trading capital to its algorithm, Rates Management Trading (“RMT”).

RMT was designed to replicate the risk profile and profits of trades executed in BSMA. However, the algorithm’s trades were delayed, sometimes up to four days, resulting in lower profits and greater volatility in BCI. From 2011 to 2015, BCI’s capital allocation to the algorithm ranged from 17% to 52%, with the remaining capital allocation managed by live traders. The algorithm typically underperformed its own profit target by $25 million per month and underperformed the live traders tracked by the algorithm.

During this time, BCI investors were not aware that an algorithmic trading application was managing any portion of their funds, let alone a substantial amount. Additionally, BCI’s three independent directors were told an algorithmic “project” was “in the early stages of development,” when in fact, BlueCrest had already deployed it as a substitute for live traders. In 2014, BlueCrest finally disclosed the algorithm’s existence to a due diligence consultant when asked, but did not disclose that the algorithm traded with greater volatility than the live traders and resulted in lower profits for investors. By the end of 2015, BlueCrest announced it was no longer managing external client funds and returned external client capital.

Five Key Takeaways for Corporate AI and Algorithmic-Based Applications

Use of AI or Algorithmic Applications Should Be Disclosed to Clients/Customers

The SEC’s settlement with BlueCrest emphasizes that the agency’s traditional focus on disclosures will be applied to AI. Companies using algorithmic-based applications should consider whether they have obligations to disclose information in their Form ADV and prospectus about how the application is utilized and how it might impact stakeholders. The SEC alleged that BlueCrest failed to disclose: (1) that investor capital was allocated to a fund with an algorithm-directed strategy utilizing a replication algorithm for investing capital, and that this disclosure was not included in the fund’s Form ADV and prospectus; (2) the amount of fund capital allocated to the algorithm for trading; (3) that the algorithm underperformed live traders; and (4) the attendant risks of utilizing the replication algorithm-directed strategy. BlueCrest’s purported disclosures, which included limited references to “quantitative strategies,” were viewed as insufficient to alert investors and prospective investors of the fund’s use of an algorithmic trading strategy. Moreover, specific disclosures should be considered where there is potential investor confusion over the fund’s utilization of algorithmic-based investing and where the use of that strategy impacts fund performance.

In a rather non-traditional approach to disclosure, the SEC took the position that, where a fund executes a mixed investment strategy, i.e., a combination of traders and an algorithm-based application, and the AI strategy causes a significant loss, that the loss and its cause must be separately disclosed to investors. The SEC alleged that BlueCrest failed to disclose that a $28 million reduction in BCI’s profit and loss was due to RMT’s failure to capture certain FX and bond options, not due to live traders. While we are not aware of any other instances where the SEC has taken this approach to disclosure for mixed-strategy funds, this enforcement action may be an indication of the SEC’s new position regarding these types of disclosures.

Managerial and Board-Level Oversight

As is the case with many areas of company risk, managerial and board-level oversight should be carefully considered for AI applications that may impact customers, especially for companies with a fiduciary duty to those customers and for companies operating in highly regulated sectors. Executives with supervisory responsibilities may need to understand how the algorithm is being used and any related risks. Providing senior management and the board of directors with training on how the company deploys the algorithm, its potential risks, and the company’s efforts to mitigate those risks may be prudent if the algorithm poses a significant risk of causing severe harm.

BCI’s independent directors were allegedly misled by the company regarding its active use of an algorithm for external client funds. Similarly, the independent directors were unaware of the extent of allocated capital in the fund with an algorithm-directed strategy, and the conflicts of interest that arose with respect to the proprietary fund, BSMA. Obviously, the withholding of material information from independent directors can impede their ability to provide effective oversight, and can itself be an independent basis for liability under the federal securities laws.

AI or Algorithmic Applications May Need to be Assessed On an Ongoing Basis

From October 2011 through December 2015, the algorithm performed worse than the live traders. The next-day (sometimes multiple-day) trading aspect utilized by the algorithm resulted in higher execution costs and poor performance during volatile markets because of the lag in response to the market. BlueCrest did not intend for the algorithm to perfectly mimic the traders because of the expense and inefficiency that would occur if the trades were replicated in real time. An internal report on the algorithm’s first-year performance found that each day the algorithm lagged behind the live traders resulted in an 8% loss in profit.

The algorithmic-trading program was also subject to ongoing model and operational errors. Although internal management at BlueCrest was aware of the algorithm’s performance issues, no substantive changes were made to remediate and prevent issues going forward. Companies that employ algorithms in trading should consider having policies and controls in place to enable ongoing review and updates of the algorithm. The use of certain high-risk AI applications may require the designation of someone in the company that is specifically responsible for monitoring the model’s progress and results over time, and for suggesting improvements to the model—sometimes referred to as keeping a “human-in-the-loop.”

Conflicts of Interest Should Be Identified and Disclosed

Investment companies using algorithms in their business operations should also evaluate and disclose potential conflicts of interest related to allocation of investment opportunities between funds, as well as any potential impact that an algorithm-based strategy may have on the fund’s fee structure. Here, the SEC alleged that BlueCrest failed to adequately disclose in its Form ADV and prospectus: (1) the existence of a proprietary fund with a similar trading strategy to that of the fund with an algorithm-directed strategy; (2) that the fund with an algorithm-directed strategy was harmed by price movements in the market as a result of trading in the proprietary fund; and (3) that BlueCrest retained a greater percentage of its performance fees in the fund with an algorithm-directed strategy due to lower costs from the use of non-human applications as compared with the proprietary fund, which paid incentive compensation to its traders.

The SEC alleged that BlueCrest’s generic disclosure that it “may” manage proprietary funds did not adequately alert investors in BCI that it was actually managing a proprietary fund simultaneously. As we have seen with other recent SEC enforcement actions addressing inadequate disclosures, “may” disclosures are viewed as insufficient where a practice is in fact occurring.

New Regulations Are Not Needed for AI-Related Enforcement Actions

This recent action is one of many signs of increased regulatory scrutiny of AI and algorithmic-based applications in the coming months. Given that the alleged conduct spanned from 2011 to 2015—and that BlueCrest is no longer a registered investment advisor—this case may be read as a signal that the SEC intends to pay closer attention to the controls and processes in place that monitor and update AI, and to assessing whether investment advisers adequately disclose the use of AI or complex algorithms to investors who may be impacted.

Moreover, this settlement demonstrates that new regulations are not required for the SEC and other regulators to bring enforcement actions related to the use of complex models. Here, the SEC used general fiduciary principles and disclosure obligations under the Investment Advisers Act of 1940 to crack down on a company’s algorithmic trading. Regulators seem very willing to use existing laws and regulations to bring enforcement actions in connection with the use of AI. Accordingly, companies should consider how current regulations may apply to their use of AI, because regulators are not waiting for new legislation to bring enforcement actions when they see AI uses that they dislike.

To subscribe to the Data Blog, please click here.

The authors would like to thank Debevoise law clerk Tricia Reville for her contribution to this article.

Author

Avi Gesser is a Debevoise cybersecurity and litigation partner. He is a member of the Debevoise Data Strategy & Security Group, as well as the White Collar & Regulatory Defense Group. Avi has extensive experience advising on a wide range of cybersecurity matters, incident response issues, data strategy concerns and artificial intelligence risks. He can be reached at agesser@debevoise.com.

Author

Robert Kaplan is a litigation partner based in the firm’s Washington, D.C. office. He has significant experience with a broad range of securities-related enforcement and compliance issues, including those involving requirements affecting SEC-registered investment advisers affiliated with hedge funds, private equity funds, investment companies, mutual funds and separately managed accounts. He can be reached at rbkaplan@debevoise.com.

Author

James Amler is a counsel in Debevoise's White Collar & Regulatory Group, Investment Management Litigation Group and FinTech Group. He can be reached at jbamler@debevoise.com.

Author

Anna R. Gressel is a senior commercial litigation associate at Debevoise and a member of the firm’s Technology, Media & Telecommunications and Data Strategy & Security groups. She actively advises clients on the legal and regulatory implications of artificial intelligence (“AI”) and other emerging technologies. Her practice includes not only representing companies in regulatory inquiries concerning AI, but also assisting companies in developing AI principles and governance mechanisms. She can be reached at argressel@debevoise.com.

Author

Steven Tegrar is an associate in Debevoise's Litigation Department. He can be reached at sgtegrar@debevoise.com.

Author

Valerie A. Zuckerman is an associate in the Litigation Department. She can be reached at vazucker@debevoise.com.