AI-Powered Crypto Scams Outpace Forensics, Costing Billions in 2025
Crypto forensics has made significant advancements, with platforms like Chainalysis, TRM Labs, and Elliptic successfully recovering an estimated $34 billion in illicit funds. These sophisticated tools are now standard practice for over 45 regulators worldwide, bolstering efforts to trace stolen money and attribute criminal activities across the blockchain, even as AI crypto scams rise.
Yet, the landscape of digital asset crime is shifting dramatically. AI-powered scams have surged in 2025 and 2026, proving to be far more profitable and scalable than traditional cons, often outpacing the real-time detection capabilities of existing forensic methods. This escalation led to estimated crypto scam losses of $17 billion in 2025 alone.
AI crypto scams present new challenges for forensics
Blockchain forensic platforms have certainly evolved, moving beyond mere reactive investigation. Today, newer generations of these tools incorporate artificial intelligence to offer predictive capabilities, scoring wallet behavior against more than 50 features and retraining daily to identify suspicious patterns.
One vendor boasts a 98% accuracy score across 14 million wallets, showcasing the precision now achievable. Rug-pull scanners, integrated into AI trading agents, can assess crucial indicators like liquidity locks, freeze authority, and deployer history in about five seconds. One service reported scanning over 881,000 token addresses and flagging 271,000 as high-risk, illustrating their enhanced ability to preemptively identify threats.
And it’s not just about active monitoring. Advanced wallet-clustering tools can spot a “sleeper” address that lay dormant for years, only to become active immediately before a liquidation event. This development represents a crucial step in understanding complex fraud schemes before they fully unfold. Regulators worldwide are increasingly adopting these technologies to maintain market integrity.
Escalating Financial Toll of AI Crypto Fraud
Despite these defensive advancements, AI-powered scams have driven staggering losses. Chainalysis reported total crypto scam and fraud-related losses reached roughly $17 billion in 2025, a sharp increase from $9.9 billion the previous year. The Federal Bureau of Investigation (FBI) reported that US crypto fraud amounted to $11.36 billion in 2025, marking a 22% year-on-year jump.
What truly changed the game for investors and due diligence operators is Chainalysis’ finding that AI-powered scams were 4.5 times more profitable than traditional cons. AI enables fraudsters to automate the creation of fake support agents, investor personas, and trusted insiders at an unprecedented scale, making scams faster to deploy and more convincing.
TRM Labs observed a roughly 500% increase in AI-enabled scam activity over the past year, leading up to February 23, 2026.
Lior Aizik, co-founder and Chief Operating Officer at crypto exchange XBO, has publicly warned about this trend. He notes that impersonation scams are increasing and becoming far more sophisticated industry-wide.
This type of fraud, where criminals pose as legitimate entities like banks, investors, or crypto influencers, posted an alarming 1,400% year-on-year growth between 2024 and 2025. This focus on targeted, personalized cons, rather than mass “spray-and-pray” tactics, pushed the average payment size from $782 in 2024 to $2,764 in 2025, a 253% increase.
Why Predictive Defense Falls Short
The core challenge stems from the inherent nature of forensic tools: they’re primarily designed for detective work, not genuine prediction. An investigation typically requires a crime to have already been committed, leaving a victim and a trail of lost funds. Even advanced predictive models are trained on historical data, meaning they learn from yesterday’s scams while tomorrow’s attackers rapidly innovate new methods.
This dynamic was starkly illustrated by the FBI’s NexFundAI sting operation. Federal agents created a fake honeypot token to catch wash traders, an operation that led to arrests announced by the Department of Justice.
However, within a day of the DOJ’s announcement, someone cloned the exact smart contract and launched a copycat token, netting its creators $127,000 in a single day using the very tactics the FBI had just exposed.
Every public disclosure meant to aid defenders can inadvertently provide a blueprint for attackers. Fraudsters often move faster to exploit these insights than regulators can implement new patches or preventative measures. This asymmetry creates a continuous arms race where the advantage frequently goes to the first mover.
Attackers Exploit Speed and Attention Gaps
The low effort required for modern attacks, contrasted with complex defensive measures, highlights another critical vulnerability. Software developer Peter Steinberger built a popular open-source project that allows an AI assistant full system access via apps like Telegram, WhatsApp, and Discord. The product required a rebrand after a trademark dispute.
Within minutes of the rebrand announcement, someone hijacked his old GitHub and X accounts. They used these compromised profiles to launch and pump a token that quickly reached a $16 million market cap before crashing by over 90%.
This exploit didn’t involve malware or stolen keys; it simply capitalized on a momentary gap in attention that no forensic tool was actively monitoring, as nothing illegal had technically occurred until the pump and dump began.
AI Agents Vulnerable to Smart Contract Exploits
It’s not just human investors falling prey to these evolving tactics; the very AI agents designed to protect and trade on their behalf are also vulnerable. Many pitches for agent-based funds promise automated, secure trading, but these agents can lose money just as easily as their human counterparts. The issue often lies in a failure to properly integrate robust security checks into the decision-making process.
A developer recently described how an AI agent on the Solana blockchain purchased a token that subsequently rugged by 94% within twenty minutes, costing the agent’s wallet $12,000.
Investigation revealed the token had freeze authority enabled, the top 10 holders controlled 91% of its supply, and the deployer had a history of launching three prior scam tokens.
All of these red flags should have been instantly detectable by current forensic tools, but the agent failed to check, seeing only a price and buying without a robust safety layer.
This highlights a critical failure mode that investors and operators must now stress-test in any agent-based fund pitch. The promise of automated trading is compelling, but the underlying mechanisms must prioritize security checks above all else. Brian Armstrong, CEO of Coinbase, warns that finance must move on-chain to avoid obsolescence, emphasizing the need for robust security as digital transactions become more prevalent.
The Blurring Lines of Trust and Deception
Perhaps the most insidious aspect of this new wave of fraud is that some of the most damaging scams never even touch a smart contract until it’s too late. AI has become incredibly adept at manufacturing false trust, creating scenarios where victims are defrauded before any on-chain transaction occurs that forensic tools could flag.
These schemes exploit human psychology, leveraging advanced AI to bypass traditional security layers.
In May, for example, a woman in Guelph, Ontario, lost $14,000 after believing she was speaking with popular YouTuber Mr Beast about a crypto investment. She was, in fact, interacting with an AI-generated impersonator. Mr Beast has been battling AI deepfakes using his likeness to push fake giveaways for years.
Other prominent figures like Elon Musk, Bill Gates, Mark Zuckerberg, Vitalik Buterin, and Mark Cuban have also been subjected to deepfake videos promoting fraudulent crypto projects or giveaways on platforms like TikTok and X. One deepfake Elon Musk video, streamed live on YouTube between March 2024 and January 2025, collected at least $5 million.
Key details
Scammers also deploy AI for sophisticated phishing emails, creating fake websites and personalized direct messages tailored to a user’s online behavior. Voice cloning technology has also advanced significantly, with Fortune reporting it has crossed an “indistinguishable threshold.”
This enables criminals to replicate a person’s voice using as little as 3-10 seconds of audio, often from social media or voicemails, to impersonate family or friends in distress. This allows fraudsters to impersonate family or friends in distress.
These frauds frequently occur in video calls, chat applications, or through seemingly legitimate customer support, exploiting a moment of trust long before any transaction is recorded on the blockchain.
By the time an analytical platform has a transaction to score, the victim’s decision to part with their funds has already been made, placing the fraud squarely in the realm of off-chain human vulnerability. As crypto tokens surge and markets become more volatile, the human element becomes an even more critical vector for attack.
The Ongoing Battle Between Offense and Defense
In this escalating arms race between crypto forensics and AI crypto scams, it’s fair to say neither side holds a definitive lead. The advancements in forensic and predictive tools are real; billions in illicit funds have been frozen or recovered, and dismissing these successes would be disingenuous.
But “real and improving” isn’t the same as “ahead.” The data from 2025 clearly indicates that, in terms of sheer dollar value, the offensive capabilities have improved at a faster rate than the defensive ones.
The primary reason for this asymmetry is straightforward: detection tools typically answer the question “is this wallet suspicious?” — but that question is only posed after someone has decided to check.
Key details
And then there are the growing number of cases, like the Guelph victim, where there isn’t even a wallet to scan in the first place, as the initial fraud is orchestrated entirely off-chain. AI has made these trust-based cons more prevalent and harder to detect proactively.
Consequently, for those involved in evaluating new projects or technologies, AI has transitioned from being a mere selling point to a critical feature that requires rigorous stress-testing.
While the blockchain provides an immutable record of wallet history, it remains powerless to verify the authenticity of a phone call or a video interaction, leaving a significant gap for AI-powered deception to exploit.