NATIONAL LAW UNIVERSITY ODISHA
ADVANCED CORPORATE LAW
A CRITICAL ANALYSIS OF SEBI’S AI/ML CONSULTATION PAPER
Under the Guidance of:
Ms. Kaushiki Brahma
Submitted by:
Aryan Chowdhury (2023/BALLB/020)
Mridula Singh (2023/BBALLB/056)
C ASELAW A NALYSIS : I NDIA AND A BROAD
Although, the jurisprudence surrounding the use of artificial intelligence and machine
learning is rather recent, and honestly, judicial developments around it few and far, a look at
the cases which do discuss the same is impertinent to understand how the Courts may
adjudicate future cases of AI manipulation or misuse in the markets. The analysis provided
forthwith does not only include Indian pronouncements (which are rather insignificant in its
binding value and few) but also cases from other jurisdictions where such cases did come
before the Courts and were later adjudicated upon.
INDIA
SEBI vs. OPG Securities Pvt. Ltd. & Ors. 1 is a landmark caselaw which established the
principle that gaining an unfair technological advantage through the exploitation of market
infrastructure constitutes a fraudulent and unfair trade practice. OPG Securities had misused
NSE’s co-location servers to conduct trades earlier than its competitors, by taking advantage
of its emergency servers, which had a significantly lower load than its main server. The
Hon’ble SAT held that such conduct was a clear departure from the principles of fair,
equitable, and transparent trade that underpin market integrity. Although not connected to
algorithmic trading or AI, the case showcases how technological growth and increased use of
assistive tech can make gaining unfair advantages easier.
In Vilas Prabhakar Lad vs. Unique Identification Authority of India and Ors. 2, the Hon’ble
Bombay High Court, in its orbiter, noted the potential misuses of AI. Although the case
revolved around a case of identity theft, the Division Bench noted that artificial intelligence
can be misused in financial markets to create fake trading accounts, and used by firms and/or
‘ill-minded individuals’ to profit off the market or manipulate the market to the benefit of
some and the loss of genuine traders.
Moreover, in Punit Trade Resources LLP vs. The Assistant Commissioner Income Tax, Circle
22(1) and Ors.3, the Bombay High Court had previously held that artificial intelligence has
been increasingly adopted by statutory bodies like SEBI to optimize transactions and make
paperwork easier for stakeholders viz-a-viz trading entities.
1
SAT Appeal No. 93 of 2019
2
2025:BHC-OS:11428-DB
3
2024:BHC-OS:10086-DB
2
In Re: Ashok Maheshwari and Ors.4, SEBI referenced to the Supreme Court’s decision in
Tarun Kumar vs. Assistant Director Directorate of Enforcement 5, whereby the Board
observed that advancement in artificial intelligence has made money laundering easier in the
market and has left the investigative agencies to ramp up their technologies in order to meet
with the threat of technology. More importantly, it was observed that tracing the original
source and where the money has been laundered has also become increasingly difficult due to
the advent of AI.
The most recent instance is the Jane Street Investigation by SEBI, whereby SEBI temporarily
banner Jane Street India, its Indian trading subsidiary from entering the market for using an
algorithm to conduct unfair trading in the Indian options market. 6 SEBI froze the assets of
Jane Street, where Jane Street used the layering strategy (discussed below in detail) to get
massive profits from the already high-volume options market.
Hence, we see that there are some small development in the Indian capital markets regarding
the advent of AI, it is not enough to understand the full assessment of how AI can impact
trading.
OTHER JURISDICTIONS
While the cases in India do not provide much context as to how to go forward with
algorithmic trading, other nations provide with better answers.
HONG KONG
In Luk Wing Yan v CMB Wing Lung Bank Ltd7, the Court of First Instance of Hong Kong
observed that artificial intelligence, while having the capabilities of supervising banking
transactions and protecting investors from fraud, can also lead to unwanted results, especially
when banking institutions use AI to measure the health of the companies, they have invested
in. Although AI can help in the compilation of financial reports, it cannot replace human
oversight.
UNITED STATES
United States v. Coscia8 is a watershed moment in the regulation of AI trading, particularly
algorithmic trading. The case attracted much controversy in the United States and was the
4
MANU/SB/2014/2024
5
2023 INSC 1006
6
Algorithmic Trading and India’s SEBI Allegations Against Jane Street, Today’s General Counsel, 28 th August
2025, [Link]
7
[2021] HKEC 729
8
866, F.3d 782 (2017).
3
first criminal prosecution and conviction for the manipulative practice of “spoofing” under
the Dood-Frank anti-spoofing law.9 In this case, Michael Coscia had commissioned the
creation of and used two sophisticated computer algorithms to execute manipulative
strategies across the futures markets, including gold, soybean oil and foreign currencies. This
was executed in milliseconds, where the algorithms would create the appearance of
downward price pressure by placing a small, genuine order to buy a futures contract at a
slightly lower price, followed by a quick placing of several large-volume orders to sell the
same contract on the opposite side of the market at even lower prices. This would create an
impression of substantial selling interest, deceiving other traders into lowering their prices.
When the market price fell to enough to meet the initial small buy order, the order would be
fulfilled. The sell orders where then automatically cancelled by the algorithm before they
could be executed. The algorithm would then immediately reverse the process, placing a
small sell order and flooding the market with large buy orders to artificially inflate the price
and sell the previously acquired contract for a small profit. This would be repeated a thousand
times in just an hour, leading to profits of over 2 million USD.
The US Supreme Court held that while mere cancelling of orders is not punishable, however,
placing or bidding orders with the intent to cancel the same is punishable. Hence, spoofing
was held to be a manipulative tactic where orders are made with the intent to not fulfil them
at all.
Before Coscia, there was the case of Athena Capital Research LLC10 before the US Securities
and Exchange Commission which is equivalent to SEBI. Athena Capital employed “marking
the close” strategy, whereby it used a specially-made algorithm to place a massive volume of
aggressive, rapid-fire trades in the final two seconds of the market. The last-second burst of
orders was designed to overwhelm the remaining liquidity in the market and artificially push
the closing prices of thousands of NASDAQ-listed stocks either up or down. 11 The SEC
found the same to be manipulative, as it was discovered that the algorithm was trained to
execute a “crescendo at the end”, a pattern that demonstrated a clear purpose to impact the
price rather than to seek legitimate, arm’s-length execution. Moreover, the intent of the
9
Steve McNamara, United States v. Coscia as a Case of First Impression, The CLS Blue Sky Blog, 7 th
September 2025, [Link]
impression/
10
In the Matter of Athena Capital Research, LLC, Release No. 34-73369 (Oct. 16, 2014)
11
Ruben Magalhaes, S.E.C. Tells High Frequency Trading Firm Athena “Not So Fast!”, Fordham Journal of
Corporate & Financial Law, 21st January 2015, [Link]
frequency-trading-firm-athena-not-so-fast/
4
algorithm trainers was also investigated and mental intent or scienter was held to be an
important aspect in concluding criminal culpability. Hence, we can see in the US, not only is
suspicious trading behaviour by the algorithm is required to be proven but also intent.
However, this is the US model’s greatest drawback as with recent development in generative
AI and LLMs, which learn user behaviour and keep learning with every prompt, the AI might
adopt a manipulative practice such as spoofing without the coder even explicitly instructing it
to do so. This is especially the case in “black box” behaviour of AI, which are outcomes
given by an AI which even its programmer or trainer can understand.
UNITED KINGDOM
In Financial Conduct Authority v. Da Vinci Invest Ltd & Ors. 12, the same layering strategy as
in Coscia and placed orders for Contracts for Difference (CFDs), which are derivatives
whose value mirrors that of an underlying share. In a similar manner, orders were placed but
never fulfilled, while genuine trades were committed on the opposite side of the market.
Ironically, the defendants were detected by the SMARTS market surveillance system, a
supervision algorithm trained by the FCA to detect manipulative trading. Ultimately, the case,
similar to the US boiled down to intent of the traders, and they were convicted on the ground
that they knew fully well what they were doing.
Recently, in 2024, the UK FCA imposed a fine of 27.2 million USD on Citigroup Global
Markets Limited (CGML) for failures of not having detected error trades committed by a
broker, where he placed a basket order of 444 billion USD, rather than 58 million USD. 13
This led to the sale of 1.4 billion worth of US entities being sold in the US market before the
order was detected by the control systems and cancelled. This led to a sharp drop in the
market indices of several European nations. Here, the FCA bypassed the requirement of
mental scienter to prove manipulation but focused on Citibank’s inadequacy of having the
critical algorithmic systems in place to cancel such orders instantly. Instead, the system
merely sent a warning pop-up to the trader, which could be manually rejected by the trader
without even reading it, which led to order being placed in the first place.
Similarly, in 2025, the FCA fined the London Metal Exchange for not having sufficient
control systems in place to manage extreme market volatility. 14 In 2022, the nickel market
12
EWHC 2401 (Ch).
13
Press Release, FCA fines CGML £27,766,200 for failures in its trading systems and controls, Financial
Conduct Authority UK, 22nd June 2024, [Link]
14
Laura Bridgewater, FCA Fines London Metal Exchange in First Enforcement Action against RIE,
Macfarlanes, 2nd April 2025, [Link]
5
had seen losses of up to 5 million USD, of which 3.6 million USD were recovered. This was
due to the price of the nickel futures trebling in a matter of days, leading to losses for the
opters in the market. The control algorithms were found to only detect “rogue algorithmic
trading” and were not designed to ensure orderly trading during perceived or actual market
volatility.
Therefore, we see that the UK market has moved beyond merely penalising traders, also
penalising exchanges and market platforms for not having sufficient systems in place to
control the market and protect investor interests. Hence, we see that the UK system has
moved beyond rogue trading, but also working towards seeking accountability for the benefit
of the investors, especially in the options market, making sure the ethical and full use of AI
takes place in the market.
C OMPARATIVE A NALYSIS WITH IOSCO’ S G UIDELINES ON U SES OF AI IN
C APITAL M ARKETS
The primary reason for this analysis is the fact that the consultation paper mentions the
International Organisation of Securities Commission (IOSCO) papers on uses of AI (one
released in 202515 and the other in 202116) as the blueprint and inspiration for the present
consultation paper by SEBI. Through this analysis, we aim to understand to what extent has
SEBI aimed to implement IOSCO’s recommendations, what it has failed to do and how the
same can be introduced into India’s capital market to safeguard itself from the threat of AI.
At its core, SEBI’s consultation paper clearly and systematically implements the six key
measures proposed in IOSCO’s 2021 guidelines. Firstly, SEBI’s principle of Model
Governance is a comprehensive amalgamation of several IOSCO measures, directly
incorporating IOSCO’s call for designated senior management responsibility, the need for an
internal team with adequate skills and a robust over-the-loop oversight mechanism of third-
party vendors. The requisition of a senior manager with technical know-how of AI models
and the inclusion of liability of third-party vendors , SEBI mirrors IOSCO’s emphasis on
clear accountability and better internal governance frameworks.
exchange-in-first-enforcement-action-against-rie-102k786/
15
Consultation Report (CR/01/2025), Use Cases, Risks and Challenges on the use of AI in Capital Markets: An
Overview, IOSCO (2025), [Link]
16
Final Report (FR06/2021), The use of artificial intelligence and machine learning by market intermediaries
and asset managers, (2021), [Link]
6
Secondly, the paper’s Testing Framework is almost a copy-paste from the IOSCO’s
guidelines. Both mandate that algorithms be tested in a segregated environment to ensure that
they behave as expected under various market conditions. Moreover, SEBI has specifically
suggested the use of “shadow testing with live traffic”, which is a practical step to validate
performance before full deployment.
Thirdly, SEBI focuses on Investor Protection and Disclosure even more, by providing a non-
exhaustive list of operations requiring disclosure, such as algorithmic trading and advisory
services, and specifies the type of information investors should receive, including the model's
purpose, risks, and limitations. Here, SEBI goes a notch further from what the IOSCO
guidelines demanded, as IOSCO provided a suggestive list, not a mandatory one to be
followed.
Finally, Fairness and Bias principles and data quality thresholds are drawn from IOSCO’s
sixth measure, which highlights the need for high-quality, unbiased data. Both recognise that
AI model performance is fundamentally dependent on the data it is trained on, and biased
data can lead to discriminatory outcomes.
HOMEGROWN INNOVATIONS
SEBI has not merely copy-pasted IOSCO’s recommendations but also added several elements
of this own. The most notable innovation is the proposed tiered approach to regulation, with
AI being used internally subject to a less-stringent “regulatory lite” regime as opposed to AI
application which directly impact clients.
Furthermore, SEBI has directly confronted the challenges posed by Generative AI, by
outlining specific risks like malicious usage for market manipulation, concentration risk and
the potential for herding behaviour. The proposed control measures, such as watermarking AI
content, monitoring vendor concentration, and mandating circuit breakers, are a step forward
and address risks that have become much more prominent since IOSCO's initial 2021 report.
LAPSES AND AREAS FOR IMPROVEMENT
Despite the strengths of the paper, it leaves several areas untouched, especially when
compared with the 2025 report by IOSCO.
1. The regulatory lite framework does not benefit all: the proposed tiered mechanism,
while sensible in principle, may create a regulatory blind spot. The distinct of “inner”
and “outer” facing is not always clear or as easy as it seems. For example, a flawed
7
internal risk management algorithm or AI tool that generates insecure code for trading
could trigger market disruptions or harm individual investors, even without directly
interacting with the client. A more robust approach is to base the level of regulatory
scrutiny on the potential market impact and criticality of the AI applications’ uses,
rather than simply on whether it is client-facing or not.
2. Insufficient oversight for “black box” behaviour of AI: although the consultation
paper acknowledges the existence of “black box” algorithms but suggests that their
providers be registered officially as research analysts and maintain reports on the
algorithm’s logic. This is insufficient for managing the risks of models whose
decision-making processes are inherently unexplainable or the developer themselves
are unable to understand as to how the AI came with the outcome. This lack of
explainability is core risk as well according to the 2025 report. Hence, we suggest that
the focus should move from demanding an explanation for such black-box behaviour
to focusing more on continuous and verifiable testing, impact analysis and “human-in-
loop” controls for high-impact trading.
3. The need to explicitly shift from intent to accountability: As seen from the drawbacks
of the US model and the manner in which the UK model remedies the same through
prioritising accountability over intent. Even the 2025 guidelines recommend
investigating into accountability rather than searching intent, as autonomous, self-
learning AI systems may make actions which are not verified by the firm but benefits
the firm unfairly. This means holding firms liable for market harm caused by their
algorithms, regardless of intent, if they cannot demonstrate a robust, auditable, and
effective governance and risk management framework.
STEPS FOR THE FUTURE
Therefore, to better protect Indian markets and investors, SEBI could enhance its proposed
guidelines by incorporating deeper insights from global initiatives:
Refine the tiered framework by categorising AI on their potential impact on market
integrity, financial stability and investor outcomes, rather then whether the systems
are internal or external.
Black-box governance needs to be strengthened, especially for opaque models.
Investor-education on AI-generated misinformation and risks related to AI use should
be mandated.
8
Codification of accountability is important as the market participant should be fully
and legally accountable for any and all actions and outcomes of its AI systems. This
principle would incentivize firms to conduct deeper due diligence and demand greater
transparency and control from their technology partners.
C ONCLUSION
In conclusion, SEBI's consultation paper is a well-structured and necessary initiative that
successfully domesticates established international principles. It represents a critical and
timely intervention, moving to formalize oversight in an area that has, until now, operated
largely in a regulatory shadow. This proactive stance is essential for a high-growth market
like India, where the rapid adoption of AI and machine learning by financial entities presents
both immense opportunities and significant risks, particularly for a large and growing retail
investor base. By systematically aligning its core tenets of model governance, investor
protection, testing, and fairness with the foundational 2021 guidelines from IOSCO, SEBI
has signalled a clear commitment to integrating global best practices. This not only enhances
the credibility of the Indian market but also provides a degree of certainty for the
sophisticated global firms that are increasingly active within it.
By embracing these future-facing challenges, SEBI can create a truly world-class framework.
Such a framework would be dynamic and resilient, one that fosters responsible innovation by
providing clear rules of engagement, while rigorously safeguarding the integrity of India's
capital markets and protecting the interests of its millions of investors.