Despite significant advancements in blockchain forensics, AI-powered crypto scams fueled a staggering $17 billion in fraud-related losses in 2025. This figure represents a sharp increase from the $9.9 billion reported the previous year, highlighting how quickly sophisticated AI tools are being weaponized by criminals.
Blockchain analysis firms like Chainalysis, TRM Labs, and Elliptic have collectively frozen or recovered an estimated $34 billion in illicit funds, with over 45 regulators worldwide now routinely employing these tools. Yet, the rapid deployment and enhanced profitability of AI-driven schemes mean defense is struggling to keep up with offense.
AI-Powered Scams Fuel Record Crypto Losses
The numbers behind the surge in crypto fraud are stark. Chainalysis data indicates total crypto scam and fraud-related losses hit approximately $17 billion in 2025. That’s a considerable jump from the $9.9 billion recorded in 2024, showing a market where illicit activities are growing rapidly.
The Federal Bureau of Investigation (FBI) also reported substantial figures, with US crypto fraud losses reaching $11.36 billion over the same period, marking a 22% year-on-year increase. These aren’t just large numbers; they paint a picture of an escalating problem that’s taking a heavy toll on investors.
Perhaps the most concerning statistic is that AI-powered scams were 4.5 times more profitable than traditional cons. This means the same type of fraud, when enhanced with AI capabilities, yielded significantly higher returns for scammers. They’re able to create highly convincing fake support agents, bogus investors, or impersonate trusted insiders at an unprecedented scale.
This shift has also driven up the average payment size. In 2024, the average loss per scam was $782. By 2025, that figure had soared by 253% to $2,764, as criminals moved away from broad, untargeted attacks to more precise, high-value operations.
Impersonation fraud, where criminals pose as legitimate entities like banks or crypto influencers, saw an alarming 1,400% year-on-year growth. Lior Aizik, co-founder and Chief Operating Officer at crypto exchange XBO, has publicly warned about the increasing sophistication of these schemes.
Forensic Tools Advance But Face Uphill Battle
It’s not all bad news for digital asset security. The defensive side has seen genuine improvements. Modern blockchain forensics platforms are incredibly effective at tracing illicit funds, clustering wallets, and attributing entities, which is crucial for legal proceedings.
Newer generations of these tools, thanks to AI integration, are moving beyond just tracing money after it’s gone. Some predictive platforms now claim they can flag suspicious wallets before any illicit activity even occurs. These systems score behavior against over 50 features and retrain daily, with one vendor boasting a 98% accuracy rate across 14 million wallets.
There are even “rug-pull scanners” embedded in AI trading agents, capable of checking liquidity locks, freeze authority, and deployer history in around five seconds. One service reported scanning over 881,000 token addresses and identifying 271,000 as high-risk. Wallet-clustering tools can also spot dormant “sleeper” addresses that activate just before a liquidation, much like noticing someone casing your neighborhood before a break-in.
But the uncomfortable truth is that while defensive tooling has gotten dramatically better, the offensive results have improved at a faster pace. It’s a continuous rivalry, much like a generative adversarial network (GAN), where both sides push each other to evolve. And right now, the advantage often lies with the first mover in the attack space.
Speed and Accessibility Empower Attackers
A key reason better detection continues to struggle is that forensic tools are primarily built for detective work, not proactive prediction. A crime typically needs to be committed and funds lost before patterns become discernible enough for investigation. Even predictive models are trained on past scams, and attackers are quick to adapt and design new schemes based on the same publicly available information.
The FBI’s NexFundAI sting operation illustrated this stark reality. Federal agents created a fake honeypot token to catch market manipulators, seizing $25 million and leading to 18 arrests. Yet, within a day of the operation’s public disclosure, someone cloned the exact smart contract and launched a copycat token.
This new scam made $127,000 in a single day using the very tactics the FBI had just exposed, proving that every defensive disclosure can inadvertently provide a blueprint for future attacks.
Key details
The ease and speed with which attacks can be launched also present a significant challenge. Austrian software engineer Peter Steinberger, known for his open-source AI assistant project, experienced this firsthand. Following a rebrand announcement for his product, his old GitHub and X accounts were swiftly hijacked.
The perpetrator used these compromised accounts to launch and pump a token that briefly hit a $16 million market cap before crashing by over 90%.
This exploit didn’t involve malware or stolen keys; it was simply a rapid exploitation of a moment of distraction. Crucially, no forensic tool was actively monitoring for this gap because nothing illegal had happened yet. The incident underscores how quickly threat actors can capitalize on vulnerabilities that exist outside the immediate scope of on-chain analysis.
For entities concerned about the security of their digital assets, this quick-moving threat landscape means constant vigilance is essential. Understanding the market dynamics can be crucial for investors, especially when considering shifts like how Bitcoin exchange supply maintains multi-year lows, which might influence overall market sentiment.
Beyond the Blockchain: The Deepfake Threat
What’s even more concerning is that many of these damaging scams don’t even touch a smart contract until it’s too late. The fraud often happens in a less tangible space: through manufactured trust facilitated by AI-generated content.
In May 2026, a woman in Guelph, Ontario, reportedly lost $14,000 after falling victim to scammers impersonating YouTuber Mr Beast for a crypto investment. Mr Beast himself has been combating AI-generated videos using his likeness to push fake giveaways for years. These deepfake scams highlight a critical blind spot for current forensic tools.
Forensic platforms can’t flag these interactions because nothing on the blockchain happens until the money is already moving. The fraudulent decision is made during a video call or through deceptive communications, creating a false sense of security. By the time a transaction is initiated and hits the chain, the victim has already been conned.
Key details
AI is also putting automated trading agents at risk. Developers have reported instances where an AI agent on the Solana network purchased a token that subsequently rugged by 94% within 20 minutes, costing the agent’s wallet $12,000.
Investigation revealed clear red flags: freeze authority enabled, 91% of supply controlled by the top 10 holders, and a deployer linked to three previous scam tokens. Yet, the agent failed to check these basic security parameters, simply seeing a token and a price and executing a buy order.
This suggests a disconnect between the safety layers and the decision-making processes within some AI agents, a critical flaw that needs addressing for systems managing Ethereum network outlook strengthens as AI-driven DEX reports increased activity.
The Ongoing Arms Race and Future Outlook
So, who’s winning this escalating battle between crypto forensics and AI scammers? The honest answer is, neither side is definitively ahead. Both forensic and predictive tools are real and constantly improving, leading to billions in recovered funds. Dismissing these advancements simply because fraud is also growing would be disingenuous.
However, “real and improving” isn’t the same as being “ahead.” The data for 2025 unequivocally shows that in dollar terms, the offensive capabilities have outpaced the defensive ones. A significant factor here is that detection tools primarily respond to the question, “Is this wallet suspicious?” But this question is only asked after someone decides to check, and often, after a loss has occurred.
Furthermore, scenarios like the Guelph deepfake scam illustrate cases where there’s no suspicious wallet to scan in the first place. AI has amplified the frequency and sophistication of these “off-chain” frauds, shifting the attack vector.
For investors and operators, this means AI can no longer be seen purely as an advantage or a selling point; it must also be treated as a primary area for stress-testing and vulnerability assessment, requiring a more holistic approach to digital asset security.