SEC’s Motives for AI Adoption Could Strengthen Market Oversight

In August 2025, the Securities and Exchange Commission launched an internal artificial intelligence task force to try to drive innovation and operational efficiency. This move signals the agency’s transition from monitoring AI as an external trend to integrating it as an internal tool. The initiative has the potential to enhance the agency’s efficiency by using AI to enhance market surveillance, examinations, and enforcement.

Effective AI Integration

AI tools allow the SEC to bridge the gap between the agency’s constrained resources and the vast, complex, and evolving financial system it monitors.

To maximize effectiveness of AI deployment, the SEC must align specific AI tools with the unique characteristics of the data and information under review. For instance, the agency’s market surveillance analytics programs, such as the Advanced Relational Trading Enforcement Metrics Investigation System, rely on quantitative trading data for detecting unusual patterns and relationships among traders. These sizable, structured datasets are ideally suited for traditional machine learning tools, which use algorithms designed to identify anomalies across high-dimensional variables with greater speed and precision than manual oversight allows.

While traditional machine learning tools excel at analyzing structured data, generative AI and large language models are optimized to help analyze unstructured information. These tools can process vast volumes of unstructured data, such as brokerage statements, internal corporate records, and emails, to identify topical observations, semantic nuances, and qualitative patterns that traditional quantitative tools might miss.

Ultimately, the SEC may also consider deploying AI agents across its investigative workflows, including for autonomous triage of inbound leads. These AI agents can operate as digital assistants, conducting complex analyses autonomously while reducing the need for human direction. By generating, orchestrating, or running their own specialized code, AI agents can investigate intricate data patterns and perform complex calculations independently, significantly accelerating the entire investigative lifecycle.

Enforcement Can Benefit

SEC enforcement stands to benefit from generative AI’s ability to process the massive, unstructured narrative datasets—such as emails, call transcripts, internal reports—that have traditionally served as sources of potentially critical evidence. These datasets present a significant analytical challenge because they lack the consistent format best suited for database analytics. Instead, they require interpreting ambiguity, context, and semantic nuances, which challenge the limits of traditional text analytics approaches.

The most immediate application of this technology lies in the SEC’s Tips, Complaints, and Referrals, or TCR, system, which annually ingests over 45,000 leads ranging from highly detailed whistleblower reports to vague or irrelevant grievances. Managing this volume with manual review or text analytics tools creates a considerable administrative burden. Generative AI can serve as an initial triage tool by categorizing submissions based on the specificity and verifiability of their allegations, which could considerably speed up the initial review process and thereby reduce administrative burden.

Beyond simple categorizing of submissions, AI-enhanced tools can identify patterns across seemingly unrelated TCR submissions. For instance, by clustering leads that mention the same entities, individuals, or suspicious behaviors, the SEC can uncover broad market manipulation schemes that might appear as isolated incidents to a human reviewer.

This automation doesn’t replace the human investigator. Rather, it empowers SEC staff to direct their expertise toward the most complex and credible leads, which can help accelerating the timeline from the initial tip to a formal enforcement action.

More broadly, deploying generative AI can transform how SEC enforcement connects “narrative” evidence with “quantitative” market activity. In a typical insider trading or market manipulation case, investigators must link the timing of private communications directly to public price movements or trading activity.

By rapidly processing thousands of unstructured records, AI allows investigators to identify hidden connections between private communications and public trades. This can significantly compress the investigative timeline while providing the evidentiary depth required for enforcement actions.

Solutions Are Complicated

Adopting AI technology likely won’t be a straightforward solution. Before the marketplace or market participants see results, the SEC should invest more in procuring or building platforms and processes to evaluate, manage, and mitigate the high error rates still prevalent in current AI-driven processes and models.

The primary challenge lies in mitigating false positive errors (in which an AI model reports suspicious activity that isn’t present) and false negatives (in which an AI model misses actual suspicious activity). If left unmanaged, high error rates could overwhelm investigators with “noise” or, conversely, create a false sense of security. To handle these errors effectively, the SEC will require robust internal procedures to validate AI outputs before they trigger formal investigative steps.

Ensuring strict data confidentiality is another significant hurdle. The SEC can’t use public-facing AI models, as doing so would risk exposing sensitive, non-public data to external networks.

However, the marketplace now offers viable solutions for such high-security clients. These solutions often include access to powerful models within FedRAMP-certified government clouds or “on-prem” services, meaning those that allow models to run entirely within an agency’s own secure data centers, ensuring that investigative data never leaves the agency’s firewall.

Key Takeaways

The SEC’s move to formalize an AI task force marks a significant step toward a more modernized market oversight. By pairing traditional machine learning for quantitative data with generative AI for complex narrative datasets, the agency can enhance its market oversight capabilities and better fulfill its mission of investor protection.

However, realizing this potential won’t be straightforward, and challenges will include high AI error rates and strict data privacy requirements. The SEC’s success will depend on its ability to balance these powerful automated tools with robust internal procedures and secure infrastructure to better protect investors in increasingly complex markets.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.

Author Information

Ross Askanazi is senior specialist manager at Cornerstone Research and consults on commercial litigation and regulatory investigations.

Ernest Kim Song is data science principal at Cornerstone Research and has over ten years of experience consulting on all phases of commercial litigation, internal investigations, and regulatory matters.

Marina Martynova is principal at Cornerstone Research and consults on financial and economic issues arising in securities litigation and government enforcement actions.

Write for Us: Author Guidelines