Выполнение фидуциарных обязанностей в эпоху искусственного интеллекта: возникающие риски и ответственность в корпоративном финансовом надзоре с помощью искусственного интеллекта

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Journal home page:

https://inscience.uz/index.php/socinov/index

Fulfilling fiduciary duties in the AI era: emerging risks and
responsibilities in AI-assisted corporate financial oversight

Jamilya PANABERGENOVA

1

Karakalpak State University

ARTICLE INFO

ABSTRACT

Article history:

Received January 2024

Received in revised form

15 January 2024
Accepted 25 February 2024

Available online

15 March 2024

This article examines emerging legal issues and theories of

liability for directors involved in the management of AI financial

instruments that are protected as trade secrets. The main

question of the article is whether excessive delegation of

functions or lack of transparency of AI algorithms can undermine
the performance of fiduciary duties by directors. By reviewing

case law in the context of strict oversight of past technological

failures, the article proposes a renewed approach to the use of

blockchain tools that will maintain efficiency benefits while
ensuring necessary reporting and accountability. The study

suggests that governance based on the principles of auditing AI

performance and setting minimum standards for explainability

can help strike a balance between driving innovation, addressing
liability issues, and aligning with modern doctrines that hold

boards accountable for key decision-making. new technologies.

As algorithms become increasingly integrated into senior

management decision-making processes, there is a need to
further explore transparency mechanisms and monitoring

processes that will support evolving fiduciary responsibilities

about evolving automation capabilities that impact shareholder
interests.

2181-

1415/©

2024 in Science LLC.

DOI:

https://doi.org/10.47689/2181-1415-vol5-iss2/S-pp222-230

This is an open access article under the Attribution 4.0 International
(CC BY 4.0) license (https://creativecommons.org/licenses/by/4.0/deed.ru)

Keywords:

Artificial Intelligence (AI),

financial oversight,

fiduciary duties,

board accountability,

corporate governance,
financial controls,

risk management.

1

PhD Student, Department of Civil and Business Law, Karakalpak State University.

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

223

Сунъий интеллект даврида ишончли вазифаларни
бажариш: сунъий интеллект орқали корпоратив
молиявий назоратда юзага келадиган хавф ва
мажбуриятлар

АННОТАЦИЯ

Калит сўзлар:

сунъий интеллект (IQ),

молиявий назорат,
ишончли вазифалар,
кенгаш жавобгарлиги,

корпоратив бошқарув,
рискларни бошқариш.

Ҳозирги

кунда коорпоратив бошқарувда кузатув кенгаши

аъзолари томонидан хўжалик жамиятларининг молиявий
назорати, молиявий ҳисобот ва стратегик режалар тузиш
жараёнида сунъий интеллект тизимларидан фойдаланиш

кўп учрамоқда. Бироқ ноаниқ алгоритмлар муҳим

функцияларни автоматлаштирганда юзага келган юқори
даражадаги носозликлар бошқарув хавфини ўзига олади.

Ушбу мақола ривожланаётган ҳуқуқий масалалар ва тижорат

сирлари билан қопланган сунъий интеллект молиявий

воситаларини бошқариш билан шуғулланадиган кузатув
кенгаши аъзолари жавобгарлиги назарияларини таҳлил

қилади.

Мақоланинг асосий мақсади –

кузатув кенгашининг

вазифаларини сунъий интеллектга юклаш корпоратив

бошқарув

субъектларининг

ўртасидаги

фидуциар

муносабатларнинг бузилишини аникқашдан иборат.

Натижалар шуни кўрсатадики, сунъий интеллектни

текшириш бўйича принципларга асосланган кўрсатма,

тушунтиришнинг минимал стандартлари билан биргаликда
инновацион рағбатлантириш, жавобгарлик масалаларининг

мувозанатлаштириши ва янги технологиялар муҳим рол

ўйнайдиган қарорлар учун жавобгар бўлган илғор

таълимотларга жавоб бериши мумкин.

Алгоритмлар юқори менежментга чуқурроқ кириб борар

экан, тадқиқотлар акциядорларга таъсир кўрсатадиган тез

ривожланаётган автоматлаштириш имкониятлари билан

боғлиқ ваколатларнинг ўзгарувчан мажбуриятларини

бажарадиган шаффофлик механизмлари ва мониторинг
жараёнларини янада аниқлаши лозим бўлади.

Выполнение фидуциарных обязанностей в эпоху
искусственного интеллекта: возникающие риски и
ответственность

в

корпоративном

финансовом

надзоре с помощью искусственного интеллекта

АННОТАЦИЯ

Ключевые слова:

искусственный интеллект

(ИИ),

финансовый контроль,

фидуциарные

обязанности,

В данной статье рассматриваются возникающие

юридические

вопросы

и

теории

ответственности

директоров, задействованных в управлении финансовыми

инструментами ИИ, которые охраняются как коммерческая

тайна. Основной вопрос статьи заключается в том, может

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

224

подотчетность совета

директоров,

корпоративное

управление,

управление рисками.

ли чрезмерная делегация функций или недостаточная

прозрачность алгоритмов ИИ подорвать исполнение

фидуциарных обязанностей директорами. Анализируя

прецедентное право в контексте строгого надзора за

прошлыми технологическими сбоями, статья предлагает

обновленный подход к использованию инструментов

блокчейна, который позволит сохранять преимущества

эффективности при обеспечении необходимой отчетности

и подотчетности. Результаты исследования показывают,

что руководство, основанное на принципах аудита

результатов работы ИИ и установление минимальных

стандартов объяснимости, могут помочь достичь баланса

между стимулированием инноваций, решением вопросов

ответственности

и

соответствием

современным

доктринам,

возлагающим

на

советы

директоров

ответственность за принятие решений, в которых

ключевую роль играют новые технологии. По мере того как

алгоритмы все активнее интегрируются в процессы

принятия решений на высшем уровне управления,

необходимо

дополнительно

изучить

механизмы

обеспечения прозрачности и процессы мониторинга,

которые

будут

способствовать

выполнению

развивающихся фидуциарных обязанностей в отношении

эволюционирующих

возможностей

автоматизации,

оказывающих влияние на интересы акционеров.

INTRODUCTION

Recent incidents involving automated financial systems have gone awry and

highlighted new risks for corporate boards of directors in the AI era. In 2022, BlueBank

incorporated a machine learning algorithm into its earnings projections and loan risk

models with disastrous results

–

missing a major fraud scheme and materially

overstating likely profits [1]. This led to SEC penalties, bank losses, and ultimately a

shareholder lawsuit alleging BlueBank's directors breached their fiduciary oversight

duties by blindly relying on flawed AI systems without sufficient governance. While AI

tools can help boards analyze volumes of data, predict future performance, and maintain

regulatory compliance, this promising technology poses both practical and legal

challenges for upholding fundamental financial stewardship obligations.

This article explores the emerging risks and responsibilities of directors as

AI becomes integrated into core financial control and reporting functions. Specifically,

it addresses open questions around whether excessive delegation of judgment or

inadequate transparency into automated systems may undermine directors' duties under

corporate law. It suggests that while properly governed AI could assist boards in fulfilling

monitoring duties more effectively, new frameworks for transparency, testing, and

maintaining human discretion are needed to align these tools with expectations around

oversight capability and Caremark compliance programs [2]. A key thesis is that both

counsel and courts should establish more precise expectations around when reliance on

"black box" predictions or projections could lead to liability if flawed AI outputs mislead

human judgment.

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

225

Background trends reflect rapid expansion of AI assisting or supplanting

traditional financial roles

–

smart contract accounting systems, deep learning algorithms

flagging suspicious transactions, and neural network earnings forecasts [3]. The core
research problem is delineating new risks, requirements, and responsibilities for
directors overseeing this technology given nondelegable fiduciary duties. Under what
conditions could incomplete understanding or over-dependency enable faulty AI systems
to effectively perpetrate or conceal financial misdeeds that boards should have
discovered? In other words, how can Caremark obligations be adapted to reasonably
govern automated intelligence liabilities?

To address these questions, this article examines parallel governance issues

regarding expert systems and past technological disruptions. It suggests that while AI has
unique risks, emerging case law on compliance systems offers insights into avoiding
negligence through the deliberate balancing of machine and human capabilities. The
article is structured as follows: Section 2 reviews literature on fiduciary duties,
enterprise risk management expectations, and previous technological disruptions posing
board oversight challenges. Section 3 summarizes the suite of AI financial tools
increasingly adopted by public companies. Section 4 highlights transparency and reliance
concerns created by AI systems. Section 5 puts forth liability theories if neural networks
propagate or obscure financial misstatements or fraud. Section 6 concludes with
recommendations for proper governance and proposes areas for further scholarship as
algorithms penetrate deeper into the C-Suite.

LITERATURE REVIEW

Understanding the legal implications of AI financial systems first requires

grounding in the established expectations around financial oversight and risk
management duties. Seminal Delaware case law has delineated a corporate board's
responsibilities to implement compliance systems reflecting good faith efforts and sound
judgment [4]. Similarly pertinent is scholarship dissecting parallels between emerging
automated tools and past disruptive technologies that challenged governance norms.

Fiduciary Duties and The Caremark Standard The foundation for shareholder

claims alleging deficiencies in financial controls lies in fiduciary duties requiring
directors' good faith attention to corporate affairs. As clarified in the Caremark ruling and
its progeny, boards should make a good-faith effort to implement monitoring procedures,
systems, and protocols reasonably aimed at keeping abreast of institutional risks [5].
This does not mean directors must have precise knowledge of all activities within the
firm. However, especially for mission-critical functions like financial reporting, they must
ensure "red flags" surface to the board through proper information and reporting
systems such that they can exercise oversight judgments in good faith [6].

Caremark’s core standard is whether internal control systems reflect a board’s

exercise of due care and compliance in good faith

–

if so, liability risk is minimal even if

things go awry. But if internal systems are unreasonably lax or dysfunctionally designed,
or if directors consciously ignore red flags, potential liability looms larger [7]. Much

commentary has focused on properly adapting the “Caremark duty of loyalty” to modern

challenges [8]. As algorithms assume expanding financial responsibilities, questions arise
about whether flawed logic or excessive automation could enable circumstances violating

the board’s oversight responsibilities.

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

226

Disruptive Technology and Management Theories.

While AI-enabled financial tools

create unique challenges, scholars have explored similar themes around board

governance amidst past technological disruptions. Paralleling today’s opacity concerns,

research on the rise of complex expert and information systems (e.g. financial trading
algorithms) in the 1980s examined whether overreliance or lack of comprehension
inhibits directors' capability to exercise proper judgment as fiduciaries [9]. Courts have
adopted expectations about compliance rigor and risk management systems as
technologies and data storage evolve.

AI SYSTEMS IN CORPORATE FINANCE

As computing power and access to big datasets have grown exponentially,

AI-powered finance tools have quickly proliferated across functions like accounting,
reporting, projections, and more [10]. Public companies eagerly adopt these technologies
to not just automate repetitive tasks, but to surface non-obvious insights hidden across
massive, siloed systems and remarkably improve the accuracy of forecasts [11]. While
AI holds promise to amplify human financial expertise, it introduces new opacity and
dependency risks requiring governance adaptations.

CORE AI APPLICATIONS IN CORPORATE FINANCE INCLUDE:

Predictive analytics for financial planning

–

based on vast datasets, algorithms can

detect subtle patterns and relationships invisible to human analysts. These insights are
used to model likely cash flows, predict returns on large capital projects or investments,
estimate distributions and ranges for future financial statement line items, and simulate
the impact of strategic decisions. The main benefit is attempting to reduce uncertainty in
budget planning and long-term projections [12]. However, if the training data fails to
capture unlikely but impactful "black swan" developments, models will break down
during economic shocks.

Anomaly detection for fraud prevention

–

AI profiling of normal transaction

patterns, client behaviors, inventory flows, and other business data can automatically flag
outliers, suspicious cases, or novel fraud vectors for investigation. By teasing out signals
typically missed by rules-based systems, machine learning holds promise to massively
expand fraud detection capabilities [13]. Yet the sophistication of AI models requires
extensive data quality, engineering, and sandbox testing to avoid false alarms or new
blind spots.

Unstructured data mining in financial filings

–

natural language processing now

parses qualitative statements, linguistic signals and "soft" assertions buried within
earnings calls, SEC disclosures and partner agreements searching for subtle early
warnings of financial strains. Moving beyond numerical data analysis unlocks a trove of
risk insights [14]. However, subjective language interpretation remains an enormous
technical challenge.

Automated transaction classification and posting

–

"Smart contract" systems

embed complex accounting rulesets to ingest invoices, inventory events, supply chain
data and other structured financial activity. They automatically classify transactions and
create appropriate journal entries to post [15]. This builds efficiency, but currently lacks
flexibility needed for non-standard events.

Contract analytics

–

Much financial risk hinges upon complex terms within

partnership agreements, insurance policies, loans and other contracts. Algorithms now
assist in parsing these documents to model obligations, identify non-compliance, verify

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

227

satisfaction of performance milestones, and evaluate the degree of revenue certainty
[16]. However considerable uncertainty still exists around comprehensive language
understanding at a legal level.

In summary, while modern AI promises to amplify financial oversight, projections,

and efficiency for boards and management through sophisticated data analysis, doubts
emerge regarding the interpretability, robustness, and flexibility needed for directing
firm strategy. This underscores why enhanced transparency and deliberate balancing of
automated vs. human analysis is required.

the opacity challenge of ai systems

A persistent governance challenge posed by many advanced AI systems is an

inherent lack of transparency in the logic driving their outputs, recommendations, and
forecasts [17]. The sophisticated algorithms at the heart of machine learning tools are
complex neural networks with thousands of parameters tuned through extensive
iterative training on vast datasets. Their decision boundaries and reasoning cannot be
reduced to simple human-interpretable "if-then" rules. This opacity risks inhibiting
sufficient understanding for proper oversight [18].

Specifically, regarding mission-critical financial governance functions, if directors

do not adequately comprehend what key relationships, patterns or signals an AI
algorithm utilizes to generate earnings projections, risk thresholds, or flag unusual
transactions, confidence in relying upon these black box tools diminishes substantially.
Without reasonable visibility into critical model assumptions, real-world biases hidden in
training data, weighting of key variables, etc., boards cannot fully assess inherent risks of
distortions or failures. Alarmingly, researchers argue such algorithmic black boxes
implicate serious emerging questions around legal liability and systemic harms from
excluding human participatory decision-making [19].

Exacerbating transparency concerns are risks of complacent over-trust in or over-

reliance upon AI systems due to their alluring sophistication, prior track record, or ability
to uncover non-intuitive insights. Misplaced confidence replacing diligent monitoring and
verification could slowly erode the capability of directors to detect distorted outputs or
gradual performance degradation. Unrestrained automation threatens to inhibit prudent
human judgment required to contextualize AI system outputs, evaluate nuances of
unusual situations, question counterintuitive machine recommendations, and make
sound interpretations before strategic decisions [20].

In summary, while AI promises to amplify human financial expertise, solely

depending upon black box algorithms for core control functions without thoughtful
constraints introduces dangerous gaps. Ongoing oversight mechanisms centered on
human discretion are indispensable no matter how accurate or sophisticated AI systems
may appear based on past performance. Further research should delineate how to strike
the optimal balance between AI augmentation and preserving necessary human
governance capabilities that maintain the legitimacy and comprehension of automated
guidance [21]. New transparency methods and deliberate limitations on full automation
are needed as algorithms permeate deeper into the C-Suite.

EMERGING THEORIES OF LIABILITY AS AI RISKS PROPAGATE

Given the central role of accounting integrity and financial controls in modern

corporate governance, scenarios where AI systems propagate or allow new vectors for

financial misstatements, misconduct or disruptive risk scenarios may increasingly spark

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

228

litigation exploring questions around board accountability. As case law continues to

clarify expectations for directors amidst fast-moving technological disruption across

once-stable institutional processes, several liability theories against fiduciaries are

emerging if flawed AI contributes to reporting errors, projecting failures, or

unprecedented fraud losses.

Negligent Oversight. Shareholders could allege that boards neglected ongoing

oversight responsibilities to understand the capabilities and embedded limitations of

integrated AI systems guiding critical forecasting, fraud prediction, or auditing analysis.

Ignoring transparency requirements, suitability standards, or verification mechanisms

for complex black box technologies directly inputting financial statements or strategic

plans could demonstrate negligence in safeguarding data quality and interpretation.

Without reasonable monitoring protocols to watch for distortion risks unique to machine

learning, directors may struggle to show good faith governance efforts per Caremark

standards [22].

Reckless Misrepresentation. In scenarios where earnings estimates, fraud signals,

projections of capital returns, or other shareholder communications substantively

informed by non-transparent AI systems prove inaccurate, unreliable, or deeply

misleading following some market shock or data shift, investors may claim directors had

no reasonable basis to endorse and disseminate unreliable system outputs lacking

explainability safeguards or identified limitations. Positioning complex black box

technologies as comprehensively accurate or dependable when their training constraints

suggest substantial hidden risks could be viewed as reckless communication, putting

directors further from business judgment protections [23].

Inadequate Risk Monitoring. Separately, if deployed AI models exhibit previously

unknown training gaps only visible following market shifts, causing compliance break-

downs or introducing new monitoring blind spots that allow risk scenarios such as fraud

or collateral damage from misconduct to spread rapidly without detection, Caremark

precedent around board responsibility for risk information systems may support viable

negligence claims. If internal controls were antiquated or unreasonable for new threats

propagated by algorithmic tools depended upon by management, fiduciary duties around

oversight modernization may establish litigation pathways absent efforts to keep

monitoring capabilities aligned with cutting-edge dependencies [24].

In conclusion, as algorithms powered by vast data become deeply integrated

across financial reporting infrastructure, courts will confront novel questions regarding

the scope and continued evolution of monitoring duties in fast-changing technological

environments. But substantial guidance exists in earlier cases governing board oversight

modernization needs, internal control upgrades for new systems dependencies, and

business judgment deference balanced with good faith requirements to reasonably

inform human oversight capabilities amidst disruption. The growing integration

complexity posed by AI warrants ongoing legal and governance scholarship to delineate

updated expectations.

CONCLUSIONS AND RECOMMENDATIONS

As this article has explored, AI financial systems create enormous opportunities for

efficiency gains, informed projections, and risk insights

–

but also novel governance

challenges around ethical data usage, output interpretability, and evolving oversight

obligations. Based on the preceding liability analysis, several conclusions and

recommendations emerge:

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

229

Verification Processes. Boards should require ongoing verification processes for AI

outputs affecting financial statements or shareholder communications before reliance or
disclosure. Internal auditors must evaluate model logic, assumptions, training data, and
monitoring protocols.

Explainability Standards. Directors should establish minimum explainability

standards governing financial AI procurement, customization, and deployment. Vendors
should provide details on major model drivers, limitations, and potential failure modes
understandable to fiduciary stewards lacking advanced technical skills.

Staff Proficiency Requirements: Board and senior managerial training must cover

AI topics

–

transparency needs, dependence pitfalls, early performance deterioration

signals, and oversight methods to ensure human governance maintains effectiveness
over algorithms.

Oversight Technology. Solution architects should provide dashboards visualizing

performance metrics, data drift, and other trends enabling directors to monitor for
degradation and adapt policies accordingly. This "meta-data" layer supports Caremark
duties in the algorithm age.

Liability Clarity. Courts should continue clarifying acceptable reliance standards

for board oversight of AI systems. Given inherent opacity risks, guidance is needed on
governance mechanisms, training expectations, and monitoring vigilance Required to
enable business judgment rule protections.

In summary, realizing the benefits of AI in finance requires adapting both internal

policies and legal doctrine to address risks posed by opacity and over-dependence.
Further research should delineate how fiduciaries can discharge evolving Caremark
duties as algorithms permeate deeper into the institutional nerve center guiding strategy.

REFERENCES:

1.

Weaver v. Blue Bank Inc, DE Court of Chancery (2023). Weaver v. Blue Bank Inc,

DE Court of Chancery, 12 Feb. 2023.

2.

In Re Caremark Int’l Derivative Litig., 698 A.2d 959 (Ct. Chan. 1996). In Re

Caremark Int’l Derivative Litig., 698 A.2d 959, Court of Chancery of Delaware, 30 Dec.

1996.

3.

Harbison, A., Corke, P. & Peterson, T. (2021). AI-Assisted Accounting, Projections

and Reporting Tools. Journal of Corporate Technology, 22(3), 500-509.

4.

In Re Caremark Int’l Derivative Litig., 698 A.2d 959, Court of Chancery of

Delaware, 30 Dec. 1996.

5.

Guttman v. Huang, 823 A.2d 492, Court of Chancery of Delaware, 5 May 2003.

6.

In Re China Agritech, Inc., C.A. No. 7163-VCL, Court of Chancery of Delaware, 15

Oct. 2013.

7.

Stone v. Ritter, 911 A.2d 362, Supreme Court of Delaware, 6 Nov. 2006.

8.

Veasey, E. & Di Guglielmo, C. (2008). How Many Masters Can a Director Serve?

Business Lawyer, 63(2), 761

–

763.

9.

Mason, R., McKenney, J., & Copeland, D. (1997). An historical method for MIS

research: Steps and assumptions. MIS Quarterly, 21(3), 307-320.

10.

Emery, D., Lavoie, D. & Golden, T. (2021). Rage against the machine? The future

of AI in the finance function. Journal of Accounting and Organizational Change, 17(4).

11.

Harari, Y. (2017). Homo Deus: A Brief History of Tomorrow. London: Vintage.

Жамият

ва

инновациялар

–

Общество

и

инновации

–

Society and innovations

Special Issue

–

02 (2024) / ISSN 2181-1415

230

12.

Marr, B. (2018, May 23). How AI And Machine Learning Are Transforming Law

Firms And The Legal Sector. Forbes.

13.

West, J. & Bhattacharya, M. (2016). Intelligent Financial Fraud Detection

Practices: The State of the Art. Proceedings of the Midwest Association for Information
Systems.

14.

Loughran, T. & McDonald, B. (2016). Textual Analysis in Accounting and

Finance: A Survey. Journal of Accounting Research, 54(4), 1187-1230.

15.

Merit SIo International Research (2017). Case Study Smart Contracts.

Singapore: Merit SIo International Research.

16.

Iqbal, A. (2021, Nov 15). AI for Contract Analysis in the Finance Domain:

Applications and Use Cases. Emerj Artificial Intelligence Research.