Being an entrepreneur and investor means I’m on the other side of many playing fields. I get lineups on my desk built on roadmaps and teams that swear their appeal is real.
My job is to figure out which parts of those displays stay in touch with the blockchain. So when I tell you that the disclosure aspect of this industry has really improved, I’m not repeating the vendor’s presentation.
Blockchain forensics platforms such as String Analysis, TRM and Ellipticity Labs It has frozen or recovered an estimated $34 billion in illicit funds. More than 45 regulators worldwide now use these tools as standard practice. It helps in recovering stolen funds, which is tracked through wallet aggregation and entity attribution, and is good enough to hold up in court.
Thanks to artificial intelligence, new generations of these tools do more than just track money after it’s already been moved. Today, there are predictive platforms that claim to report on the portfolio before it ever runs.
They record behavior against over 50 features and retrain them daily. One vendor claims an accuracy score of 98% across 14 million wallets. We have scanners installed directly inside our AI trading agents, verifying liquidity locks, freezing authority and posting history in about five seconds.
One such service I mentioned It scanned more than 881,000 token addresses and marked 271,000 addresses as high risk. There are even wallet aggregators that detect a “dormant” address that has been dormant for years and only became active right before liquidation – the digital version of someone’s note that has been monitoring your street.
So, if you only read the sellers’ pages, you’ll walk away thinking that cryptocurrency fraud has basically been solved, because we now have a small army of machine learning models monitoring every chain, every wallet, and every transaction around the clock.
Then you check what the machine learning era itself did on the other side of the ledger.
The numbers behind AI crypto scams
According to Chainalogy, total cryptocurrency scams and fraud-related losses for 2025 He sits About $17 billion, up from $9.9 billion the previous year. The FBI’s figure for cryptocurrency fraud over the same period is $11.36 billion in the US alone, an increase of 22% year over year.
These are the numbers that connect to the chip of the board. But the thing that really changed the way due diligence was done was this: Chaina Analysis found that AI fraud operations were 4.5 times more profitable than traditional ones.
Same scam, same goal, but using AI, scammers can manufacture fake support agents, fake investors, or trusted people at scale.
Lior Isaac, co-founder and chief operating officer of cryptocurrency exchange XBO, has publicly warned against this Impersonation scams are on the rise And it becomes more developed at the industry level. His rule of thumb is simple: never transfer your cryptocurrencies to anyone you can’t verify, especially if the request is wrapped in urgency and secrecy.
Impersonation fraud – Criminals posing as a bank, investor or cryptocurrency influencer – achieved 1,400% year-on-year growth. Fraudsters are now using AI to carry out targeted, expensive scams on people they identify first, rather than the cheap, high-volume spray-and-pray approach they used before.
This has caused the average payment size to rise sharply, from $782 in 2024 to $2,764 in 2025, an increase of 253%. I take this personally, because investors and operators with any public profile are exactly the ones who are being cloned.
Here’s the uncomfortable part: While the defensive tools have gotten dramatically better, the offensive results have gotten better, too.
It is like a generative adversarial network, where the generator and discriminator engage in contention that continuously improves the entire model.
Both offensive and defensive tools derive from the same AI capabilities. Currently, this favors the first mover, not who builds the best model in isolation.
Why does the best detection keep losing the race?
The honest answer is that forensic tools are designed for detective work, not prediction. For an investigation to take place, a crime must have been committed.
You need a victim who has already lost money before you can trace a pattern visible enough to flag it. Even predictive models that claim to catch success before it happens are trained on yesterday’s scams — and tomorrow’s scams are designed by someone reading the same training data.
This became clear to me in real time with the FBI’s NexFundAI sting: fake token federal agents creature To catch laundry dealers.
One day after the Justice Department announced arrests linked to the operation, someone cloned the exact smart contract and released counterfeit code, generating $127,000 in a single day using the same tactics the FBI had just revealed in court documents.
Any LP asking me whether “the worst behavior in this market is finally clearing” would have had his answer within twenty-four hours.
The FBI operation became the attacker’s blueprint. Every revelation that helps the defender also provides the attacker with a working model – and attackers read faster than a patch issued by regulators.
The attack side has become cheaper and faster
You can see the same asymmetry in how little effort it takes to attack now. Software developer Peter Steinberger has built a popular open source project that lets you run an AI assistant on your computer with full access to the system via apps like Telegram, WhatsApp, and Discord.
The product had to be renamed after a trademark dispute.
Within minutes of the rebranding announcement, someone hijacked his old GitHub and Pumped a token that reached a market cap of $16 million Before it collapses by more than 90%.
No malware and no stolen keys. Just someone fast enough to exploit an attention gap that no forensic tool was monitoring. The tools were not being monitored because nothing illegal had happened yet.
When the AI agent is the one it becomes powerful
It’s not just humans falling for this that worries me, because a lot of the presentations I get are some version of “let our AI agent trade for you.” These agents can lose money on your behalf as well.
One developer described how an AI agent in Solana bought a token whose price reached 94% after twenty minutes, costing the agent’s wallet $12,000.
Upon investigation, freezing authority was enabled for the token, with the top 10 holders controlling 91% of the supply. The publisher has already released three previous scam codes.
Each of these red flags should have been checked in seconds by the detection tools I described here. But the agent did not check. She simply saw a token and a price and bought it, because the security layer was not connected to the decision layer.
This is the exact failure mode I now experience in every agent-based financing offer that crosses my desk.
The part that no tool can fix
What worries me most is that some of this damage doesn’t touch the smart contract at all. I have a public profile and following, which makes me exactly the kind of face to be cloned.
In May, it was reported that a woman in Guelph, Ontario, lost $14,000 to scammers after she thought she was talking to YouTuber Mr Beast about investing in cryptocurrencies. It wasn’t. Mr. Best has been fighting AI-generated videos that use his image to push fake gifts for years.
Forensic tools don’t identify these interactions, because there’s nothing about them that touches the chain for the money to actually move. Fraud happens on a video call, in a moment of trust. By the time there is a transaction to score on the analytics platform, the decision that will cost the victim has already been made.
AI is getting better at manufacturing that false trust faster than it is at reporting it. This is where most of the $17 billion has actually gone.
AI Crypto Scams: Who Really Wins?
Neither side.
This is the most honest answer I can give. Both sets of tools, forensic and predictive, are real. Recovery is real. To dismiss them because fraud is also on the rise would be a form of dishonesty.
But “real and improved” is not the same as “future.” The 2025 data is clear: in dollar terms, offense has improved faster than defense.
If there’s one reason for this, it’s this. Detection tools basically answer the question “Is this wallet suspicious?” – And this question is only asked after someone decides to check.
Then there are cases like Guelph, where there is no wallet to scan in the first place. AI has made these situations more common, which is why I stopped treating AI as a selling point in any presentation and started treating it as the first thing I want to stress test.
The blockchain can confirm the history of the wallet. Phone call cannot be confirmed,
this post Cryptocurrency forensics is getting smarter, but AI scammers got there first appeared first on BeInCrypto.
