AI for Market Surveillance in Crypto: Reducing Wash Trading…

Crypto markets have long faced the problem of artificial trading activity, with wash trading among the most common ways traders or market participants can make an asset appear more active than it really is. The problem is difficult to measure, because activity that looks suspicious does not always prove manipulation. The growing use of artificial intelligence (AI) is giving exchanges and regulators new ways to identify unusual trading patterns, connect activity across accounts and investigate potential market abuse.

What Is Wash Trading and How Much Does It Cost Crypto?

Wash trading occurs when a trader or group of traders buys and sells the same asset between accounts they control to create the appearance of genuine market activity. The trades can increase reported volume without creating a meaningful change in ownership or exposure.The practice can make a token appear more popular or liquid, potentially attracting genuine traders and influencing how an asset is ranked or perceived. A high trading volume alone, however, does not prove wash trading. Investigators typically look for repeated transactions, linked accounts, rapid reversals and other unusual patterns.A 2025 analysis from Chainalysis estimated up to $2.57 billion in potential wash trading across Ethereum, BNB Chain and Base during 2024. The figure came from two methods that identified $704 million and $1.87 billion in suspected activity. Chainalysis also focused on decentralized exchanges (DEXs), so the figure does not represent wash trading across the entire crypto market.Chainalysis combined two heuristics, matched buy-and-sell detection and multi-sender controller detection, to reach its $2.57 billion estimate for 2024. Data: Chainalysis 2025 Crypto Crime Report · Chart: FinanceFeeds.Chainalysis-identified suspected wash-trade volume peaked near $400M in April 2024 before declining toward year-end. Source: Chainalysis.The distinction between the two methods affects how wash trading can be identified. Centralized exchanges have access to account-level trading data and, in many cases, customer information that can help link suspicious activity to specific users. On decentralized exchanges, trading activity is visible on-chain, but the wallets behind those trades are usually pseudonymous, so investigators can track the transactions while still having to determine whether multiple wallets are controlled by the same person or group.That challenge has also produced some striking estimates over the years. A Management Science study published in 2023 found that wash trading accounted for more than 70% of reported volume on some unregulated exchanges in its historical sample, showing how significant artificial activity could become on individual platforms, although the research examined an earlier market and does not represent current conditions.A similar caveat applies to the widely cited claim that 95% of Bitcoin trading volume was fake. That figure came from a 2019 study that examined a specific group of exchanges over a short period, so it provides historical context rather than a current estimate of artificial trading volume.Anthony Georgiades, Founder and General Partner at Innovating Capital, said traditional surveillance remains useful because it can identify established patterns, while AI can examine activity across a wider range of relationships.

“Rule-based surveillance is still the foundation. It is good at identifying known behaviors: a wash-trade pattern that matches a defined threshold, repeated self-matching from the same account, an unusual cancellation rate, or trading that crosses a preset limit.”

The limitation, he said, is that manipulation can adapt to the rules designed to detect it.

“AI can look across sequences, relationships, and context rather than asking whether one event crossed one rule. It can connect activity across accounts, wallets, instruments, venues, and time periods, then identify behavior that is suspicious in combination even when each individual transaction looks ordinary.”

The scale of the problem can also extend beyond a few wallets. Chainalysis found that one controller address was responsible for approximately $142.99 million in suspected wash trading volume in January 2024, while controller addresses managed an average of 183 addresses during the year.These kind of activities shows why surveillance has moved beyond simply checking whether one account traded against itself. Research into wash trading in DeFi has also highlighted how artificial activity can spread across multiple wallets and venues.

Investor Takeaway

The Regulatory Impact On Wash Trading

Regulators have increased their focus on artificial trading activity as crypto markets have become more established. In the United States, the Securities and Exchange Commission (SEC) and the Department of Justice (DOJ) demonstrated this in October 2024 when they brought parallel civil and criminal actions against individuals and companies accused of manipulating crypto markets.The SEC charged three purported market makers, including ZM Quant, Gotbit and CLS Global, alongside nine individuals. The regulator alleged that the firms used self-trading and automated bots to create artificial trading activity, with some bots generating billions of dollars in artificial volume each day.At the same time, the DOJ announced criminal charges against 18 individuals and entities. Authorities seized more than $25 million in cryptocurrency and alleged that trading bots were responsible for millions of dollars in wash trades across around 60 cryptocurrencies. The Federal Bureau of Investigation (FBI) also created NexFundAI as part of the investigation to identify suspected market manipulators.The cases show why detecting manipulation is not simply about identifying unusual volume. Investigators need to establish who controls the accounts, how the trades are connected and whether the activity had an economic purpose.A case involving MyTrade also shows how these services could operate. Its founder, Liu Zhou, pleaded guilty after prosecutors accused the company of providing wash-trading services through automated bots. The case later resulted in a $10,000 fine, according to a recent report on the MyTrade case.The broader issue is that exchanges can have conflicting incentives when volume is commercially valuable. Kyle Reidhead, Co-Owner and Head of Research at Milk Road, argued that this makes independent oversight important.

“Reported volume is a marketing metric. It's what gets a venue ranked on the aggregators, it's what gets a token listed, it's what goes in the fundraising deck.”

That creates a difficult question for exchanges: can they be expected to police a metric that also helps attract customers and listings? Reidhead believes the answer depends partly on who is applying the pressure.

"A fund won't size a position against liquidity that might be fake, because if it's fake they can't get out of it. So for a venue serving serious capital, honest volume isn't a compliance cost, it's the product."

That incentive helps explain why institutional customers and regulators have become the important drivers of surveillance spending. The same regulatory attention is visible in US market-structure rulemaking, including a recent SEC order on how crypto assets qualify under exchange listing standards, which leans on joint SEC-CFTC interpretive guidance.

MiCA Is Making Market Surveillance A Requirement

Europe has taken a more formal approach through the Markets in Crypto-Assets Regulation (MiCA), which treats market abuse as a regulatory issue that crypto platforms must actively monitor. A point worth clarifying, raised by Jukka Blomberg, founder of NorthPoint and former chief marketing officer of two international crypto exchanges: MiCA does not use the term "wash trading" anywhere. Article 91 instead prohibits behavior that gives "false or misleading signals as to the supply of, demand for, or price of, a crypto-asset," which catches the practice by construction. The surveillance duty is narrower than often described: Article 92 binds "any person professionally arranging or executing transactions in crypto-assets," so a custody-only provider is not caught by it at all.The regulatory transition reached an important point on July 1, 2026. The European Securities and Markets Authority (ESMA) said the transitional period had ended, meaning firms relying on the old national regimes could no longer serve EU clients without the required authorization. The scale of that consolidation is significant: of the roughly 3,611 crypto-asset service providers that operated in the EU before MiCA, only around 230 had secured authorization by the deadline, according to data provider VASPnet.Of roughly 3,611 crypto-asset service providers operating in the EU before MiCA, only about 230 were authorized by the July 1, 2026 deadline. Data: VASPnet, ESMA, as reported · Chart: FinanceFeeds.The shift has also affected major industry players like Binance which withdrew its MiCA application in Greece on June 24 and said it would pursue authorization in another EU member state. The rules also make surveillance more specific: the technical standards adopted under MiCA require firms to monitor orders and transactions, generate alerts, document their systems and maintain an appropriate level of human analysis. The framework therefore creates room for AI but does not let the technology operate without oversight.Elisenda Fabrega, General Counsel at Brickken, said AI works best as an additional layer to traditional surveillance.

“Some traditional tools rely on rigid conditions or parameters, as it is for instance ‘if/then’. Once triggered, this flag trades. That layer works well for what it is designed to do, and it remains necessary.”

The difference, she explained, is that AI can look for deviations from normal behavior without requiring every possible form of manipulation to be defined in advance.

“Instead of defining the specific conduct to look for, you define what normal activity looks like and how much deviation is significant. The system then identifies what departs from that baseline without that conduct having been specified in advance.”

This approach can help identify patterns spread across several wallets, venues or accounts. It also explains why AI is becoming more useful as trading activity becomes harder to monitor through individual rules.The detailed technical standards came after the market-abuse obligations had already taken effect, while the European market still lacks a single view of trading activity across all venues. This makes cross-platform surveillance more difficult, particularly when suspicious activity moves between regulated and less regulated markets.

How AI Can Detect Wash Trading

AI can help surveillance teams detect wash trading by examining relationships that are difficult to identify through simple rules. Instead of checking whether two accounts traded with each other, a model can examine trading frequency, wallet connections, transaction timing, order behavior, and activity across different venues.This becomes especially useful when a group spreads its activity across several wallets. A trader may avoid directly buying from and selling to the same wallet while still coordinating trades through connected accounts.AI models can map these relationships and surface wash trading clusters that show unusually similar behavior or repeatedly interact with one another. Analysts can then investigate those clusters instead of manually reviewing every transaction.Jukka Blomberg explained why this matters when manipulation involves multiple accounts.

“Rule-based systems ask ‘did this known pattern occur?’ Statistical and machine-learning systems ask ‘is this account behaving unlike itself, or unlike its peers?’ The real gain is in collusive and distributed behaviour.”

He added that indirect wash trading can be difficult for conventional self-trade rules to identify because the cooperating accounts do not necessarily trade directly with each other.Anthony said this is where AI can offer an advantage over traditional systems.

“The limitation is that manipulation adapts. AI can look across sequences, relationships, and context rather than asking whether one event crossed one rule.”

The technology is already being adopted across the market. Exchanges and trading platforms have worked with surveillance providers such as Nasdaq, Eventus and Solidus Labs, while blockchain analytics firms including Chainalysis, TRM Labs and Elliptic provide tools for tracing wallet activity and identifying suspicious transactions.The adoption started before 2026, with firms such as Bitvavo adopting Nasdaq's market surveillance technology in 2024, while Bitpanda and One Trading later announced surveillance partnerships with Eventus and Solidus Labs. The growth reflects a broader move toward automated monitoring as regulatory requirements become harder for exchanges to manage manually.

AI Can Detect More Than Wash Trading

Wash trading is only one form of market abuse that AI can help identify. It can also detect potential spoofing and layering, unusual cancellation rates, coordinated buying and selling, and rapid changes in order-book liquidity.For example, a trader could place large orders to make the market appear to have strong demand and cancel them before execution. Detecting that behaviour requires the system to examine what happened to the orders over time, not simply whether a large order appeared in the book. This is why activities such as walls and spoofing can also become relevant to automated surveillance.As markets become more fragmented, AI can also connect trading activity with on-chain movements and give surveillance teams a broader view of what is happening across venues. A wallet that repeatedly receives funds from the same source before coordinated trades across several exchanges, for example, could provide another piece of the puzzle when investigators examine potentially related activity. This is important because traders can move between centralized exchanges, DEXs and other venues within seconds, creating a much larger pool of activity for surveillance teams to analyze. The same approach is now extending beyond traditional crypto trading as firms expand automated surveillance to prediction markets and tokenized assets.

Do Customers Know They Are Being Watched?

The growing use of automated surveillance raises a question that receives less attention: do customers know that exchanges are analyzing their activity? Blomberg argues that exchanges should disclose the monitoring clearly but should not treat customer consent as the legal basis for it.

"Under EU law, market-abuse surveillance is not something you ask permission for. It is necessary for compliance with a legal obligation, so consent is the wrong lawful basis, and burying 'you consent to monitoring' in the terms of service is worse than useless, because it misdescribes the relationship and implies a right to withdraw that does not exist."

Exchanges should instead explain the monitoring in their privacy notices, including why it takes place, what data is analyzed and how long records are retained.

"So: yes, it should be disclosed, but in the privacy notice rather than the T&Cs, and framed as an obligation rather than a permission. In practice most retail users have no idea, and the disclosure that does exist is written by lawyers for regulators rather than for customers."

Users do not need to know the exact rules an exchange uses to flag suspicious activity, but they should know that monitoring takes place and understand why their data is being analyzed. Privacy notices can explain the purpose of the monitoring, the types of data involved and how long the information is retained, while the platform's terms can explain what action it may take when activity is flagged.What matters to customers is what happens when that monitoring leads to action against their account. A flagged transaction or unusual trading pattern could delay a withdrawal or result in an account restriction, making human oversight especially important. Under Article 22 of the General Data Protection Regulation (GDPR), solely automated decisions that have legal or similarly significant effects are subject to additional safeguards, including provisions for human intervention in certain circumstances. Georgiades said customers should understand what happens after an alert is triggered.

“If a platform is monitoring trading behavior with automated systems, customers should receive clear, plain-language notice of that fact.”

Reidhead put the concern more simply.

“Nobody minds being watched. People mind their withdrawal being frozen for four days with no explanation.”

Overall, customers do not necessarily need to know exactly how an exchange detects manipulation, but they should understand that monitoring takes place and what process follows when their activity is flagged.

The Risk Of False Positives And False Negatives

More surveillance does not automatically mean better surveillance. AI systems can make mistakes in both directions, flagging legitimate activity as suspicious or failing to identify sophisticated manipulation.A false positive can delay a withdrawal, interrupt a trading strategy or force a compliance team to spend hours reviewing legitimate activity. A false negative can allow wash trading or other manipulation to continue and damage the integrity of a market.The problem becomes harder as trading behaviour changes. A model trained on historical examples may struggle with new manipulation techniques, while a model that is too sensitive can overwhelm analysts with alerts. There is also a shortage of publicly labeled examples of confirmed crypto market manipulation, making it harder to train and test models.Blomberg cautioned against using figures from other areas of compliance to describe trade-surveillance accuracy.

"The '90-95% of alerts are false positives' figure circulates constantly and I have never been able to trace it to a primary source, and it comes from anti-money-laundering, which is a different discipline from trade surveillance. For trade surveillance specifically, no comparable public data exists at all."

Georgiades said exchanges should therefore match their intervention to the level of risk.

“A low-confidence anomaly should generally be flagged for review, not automatically blocked. A high-confidence signal involving rapid movement of funds, coordinated accounts, or a known high-risk pattern may justify a faster intervention, but it should still generate an evidence trail and a path to human review.”

Under MiCA's technical standards, firms have to show how their monitoring works, put trained people in charge of reviewing what it flags, and maintain oversight of the whole system. The measure of a good system is not how many alerts it fires but how many lead somewhere useful, catching real abuse while leaving ordinary customers alone.

AI Can Strengthen Market Surveillance, But It Cannot Replace Accountability

The biggest limitation of AI surveillance is that it can identify unusual behavior without proving why that behavior occurred, and suspicious trading is not automatically illegal trading. A pattern can look unusual because a trader is executing a legitimate strategy, moving between venues or responding to market conditions.The Mango Markets case, involving a Solana-based DeFi trading and lending platform where trader Avraham Eisenberg was accused of manipulating the price of its MNGO token to borrow more than $110 million in crypto, illustrates how difficult that distinction can become. Trading activity can be observable and unusual, while the legal question of whether it amounts to wrongdoing still requires evidence and context.AI works best as one part of a wider surveillance process, including spotting patterns, connecting related accounts and ranking cases that deserve a closer look, while analysts examine the context and regulators determine whether the activity warrants enforcement.Exchanges sit in the same bind, because automated surveillance sharpens oversight but does nothing about the commercial pull of trading volume. A platform that profits from inflated activity still needs an independent compliance team, clear escalation steps and outside oversight. Reidhead argues that this is why surveillance ultimately becomes a regulatory issue.

"Almost nobody builds a real surveillance function because they woke up wanting one. They build it because a regulator requires it, or because a customer who answers to a regulator requires it. Take the requirement away and the spending goes away with it."

As MiCA moves into full enforcement, the public test will not simply be whether exchanges have surveillance software. It will be whether those systems consistently identify genuine market abuse, protect legitimate customers from unnecessary restrictions and provide regulators with evidence they can act on.

Investor Takeaway

Conclusion

AI is making crypto market surveillance more capable by allowing exchanges to examine relationships, trading patterns and wallet activity that traditional rules can miss. This makes it useful for detecting wash trading and other forms of coordinated manipulation across fragmented markets.AI alone, however, cannot determine whether suspected wash trading amounts to market abuse. Human review, clear regulatory requirements and accountability still matter when the system flags unusual behaviour. As crypto markets mature, the value of these tools will ultimately depend on how well they help exchanges catch manipulation while keeping legitimate trading activity moving.