Top News

By Dipal Dutta

Artificial intelligence has become a cornerstone in the fight against fraud in cryptocurrency transactions, as machine learning models examine transaction flows on blockchains in real time. They identify patterns such as sudden bursts of activity from dormant wallets or transfers linked to risky addresses. Analytics tools like Chainalysis Reactor map these flows, integrating data on mixers and darknet markets to assess risk even when individual transfers seem innocuous. This real-time analysis facilitates immediate action, flagging suspect activity for further investigation or automatic blocking.

Fraud rings these days themselves use AI to orchestrate complex scams, using deepfake imagery, voice cloning, and impersonation bots to execute phishing attacks at scale. According to Chainalysis, criminals deploy AI-generated content to automate thousands of personalised messages and fake trading platforms, enabling high-volume deception across multiple languages. In parallel, detection systems have also evolved. For instance, Chainalysis Alterya applies machine learning to scale scam detection and automate responses, continuously adapting to emerging fraud tactics.

Identifying Money Laundering

Beyond the professional space, blockchain forensics firms and universities have pioneered systems aimed at identifying money laundering on public ledgers. In 2024, Elliptic, MIT, and IBM collaborated on a model that analysed 200 million Bitcoin transactions in 122,000 labelled subgraphs tied to known illicit activity. 

Tested against real exchange data, the model flagged 52 suspicious transaction chains and correctly identified 14 accounts already suspected of money laundering, which sharply improved the detection rates from 0.1 % to over 25 % among flagged cases. Such advancements have led to going beyond identifying illicit wallets and detecting the very patterns of laundering.

The forensic tracking efforts have now extended beyond Bitcoin. Transformers like BERT4ETH are pre-trained on Ethereum smart-contract transaction sequences to detect fraud, improving phishing account identification and de-anonymisation over previous graph-based methods. More recent models, such as SmartBugBert, analyse compiled smart contract bytecode and control-flow graphs to detect vulnerabilities with over 90 % precision and recall. Large language models have also been adapted.

For instance, PonziSleuth applies zero-shot chain-of-thought prompting to Solidity code, identifying Ponzi schemes with over 96 % accuracy and minimal false negatives. Academic frameworks like iAudit combine fine-tuned detectors, reasoners, and critic agents to audit smart contracts with over 91 % accuracy. These systems assess contract logic for vulnerabilities such as transaction-ordering, access control, and timestamp dependencies for real-time contract vetting.

Defending With AI

These days, major crypto exchanges are adopting AI-driven defences. For example, Binance integrates AI into compliance and security operations, assisting specialists with law enforcement requests and freezing stolen funds. In 2023, Binance responded to over 58,000 requests within three days on average and aided in arrests involving extremist-linked funds.

The exchange also contributed to recovery efforts following DeFi rug pulls such as those involving Wine Swap. Its AI tools monitor peer-to-peer trades, detect suspicious behaviour, and enforce sanctions. In another instance, Chainalysis partnered with blockchain security firms Hexagate and Alterya to bolster real-time prevention capabilities. Hexagate brought web3 security intelligence, while Alterya added fraud detection technologies capable of identifying anomalous blockchain activity and alerting partners before harm occurs.

Chainalysis now provides Wallet Scan features that link seed phrases to wallets across chains, helping law enforcement trace illicit funds efficiently.

AI-based fraud detection and prevention is also being adopted by national regulators and banks. Only last month we saw the Commonwealth Bank of Australia introduce an AI-powered system to process financial crime alerts, create visual activity maps, and integrate generative AI for reporting.

Similarly, Google Cloud launched an AML tool in 2013 to generate risk scores from transactional patterns and KYC data, achieving up to 60 % fewer alerts and a two-to-four-fold increase in true positives for institutions like HSBC, Bradesco, and Lunar. Fintech platforms in Europe and Asia similarly blend blockchain transparency with AI predictive analytics, reducing false alarms by 60 % and preventing multimillion-dollar breaches.

Unfortunately, the research data shows that the identity theft fraud losses have risen sharply. US consumers reported $12.5 billion lost in 2024, a 25 % increase from the year before, driven by AI-enhanced deepfake and synthetic identity scams. As a result, Fraud-as-a-Service syndicates like Lazarus Group, FIN7, and Scattered Canary now operate as startups using R&D budgets and AI tools.

Financial defence now mirrors that structure, relying on traffic-level anomaly detection and behavioural analysis systems supported by cross-industry collaboration. The challenge is exacerbated by dark-web AI, which creates new vulnerabilities. Last year, Cato Networks identified tools enabling fraudsters to bypass human KYC on exchanges using AI-generated fake IDs, undermining platform safety. In response, risk teams introduce artefacts’ randomness and consistency checks to detect forgery and manipulation.

Today, AI’s power in prediction is becoming more strategic rather than brute force. For example, Coinbase has developed a model to detect exploit precursors on Layer 2 networks like Arbitrum and Optimism. The model flags wallets before any breach occurs, demonstrating that AI can anticipate attacks and inform pre-emptive hardening. National AML systems, such as those in Estonia and South Korea, use AI and blockchain analytics to track transactions in real time across decentralised systems.

Entities like DIFC and ADGM in the UAE are partnering with analytics firms to build deterrent frameworks that attract investors through transparency. At the consumer level, Google is integrating on-device AI to scan for scam messages related to crypto fraud, or "pig-butchering," processing over two billion suspicious messages monthly while preserving privacy. 

Prevention Through Prediction

Together, these developments form a layered defence model, where predictive analytics detect anomalies before exploitation, prevention tools block suspicious transactions in real time, and protection frameworks audit contract code and identity records. This ecosystem spans regulators, exchanges, banks, analytics firms, and end-user devices. Collaboration and data sharing enhance collective awareness. 

The crypto environment continues to change, prompting fraud operators to constantly update their methods. To this end, AI remains the most adaptable tool, capable of ingesting new scam types, retraining models, and automating defences at scale. It establishes a comprehensive fraud-fighting infrastructure. However, some ethical concerns remain around AI’s role in surveillance and legal use in investigations as black-box models complicate compellability in court. Moving forward, these will have to be addressed with more interpretable and glass-box type models for wider acceptance, especially among the sceptics. 

(The author is the CEO & Founder of RedoQ)

Disclaimer: The opinions, beliefs, and views expressed by the various authors and forum participants on this website are personal and do not reflect the opinions, beliefs, and views of ABP Network Pvt. Ltd. Crypto products and NFTs are unregulated and can be highly risky. There may be no regulatory recourse for any loss from such transactions. Cryptocurrency is not a legal tender and is subject to market risks. Readers are advised to seek expert advice and read offer document(s) along with related important literature on the subject carefully before making any kind of investment whatsoever. Cryptocurrency market predictions are speculative and any investment made shall be at the sole cost and risk of the readers.