- TRM Labs traced 274.60 ETH, worth about $517,205 at the time of the transfers, out of 234 contracts funded by 224 people.
- Nine near-identical YouTube tutorials pointed viewers at a fake compiler site that deployed a malicious contract instead of the code they pasted in.
- Nobody has been named, arrested or charged. TRM describes the campaign as a likely coordinated operation, based on funds landing at six shared addresses.
Nine YouTube videos promised the same thing: a trading bot, built with the help of an AI model, that would find arbitrage opportunities on its own. Following one to the end meant deploying a and putting money into it. At least 224 people did, and 274.60 ETH left their wallets.
Nothing in this required a hacked wallet, a stolen or a malicious link. Every transaction was approved by the person who lost the money, using their own wallet, doing what the video told them to do. That is why no security warning appeared at any point, and why the usual advice about phishing and dodgy approvals would not have caught it.
It also shows how AI branding is now being used as a trust signal in fraud. TRM found no AI involved anywhere in these contracts. The name was decoration.
What the videos offered
The tutorials were presented as the work of separate creators, though TRM Labs says they showed signs of common production. Between them they have collected 310,474 views since the earliest was posted. One narrator describes building a “fully autonomous crypto trading bot from scratch”.
Earlier versions of the same format used ChatGPT branding before switching to Claude. TRM examined the deployed contracts and the compiler sites and found no AI functionality, no model query and no API key. Anthropic’s technology is not part of this in any form beyond the logo.
How the money actually left
The video walks the viewer through copying the bot’s source code and compiling it. Rather than Remix, the browser-based editor most people use to deploy Ethereum contracts, the tutorial points at a lookalike site styled to resemble it. TRM found that site discards whatever is pasted into it and deploys a different contract entirely, fetched from the operator’s server.
The contract that lands carries one instruction that matters. Any balance above 0.05 ETH is forwarded to an address controlled by whoever built it, triggered the moment the user presses Start or Withdraw. The tutorials ask for 1 to 2 ETH in “startup capital”, and the median loss TRM recorded was 1 ETH, which lines up with the instruction.
Some victims were pushed for more. An on-screen message from the fake compiler reads: “ERROR: Arbitrage stuck — Gas nonce liquidity equals injected liquidity. Add 50% of your original liquidity (up to 1 ETH) as arbitrage gas to your bot’s smart contract, then press Start again.”
From there the funds moved through swaps, including into Dai, across cross-chain bridges and through at least one mixer path. TRM identified no centralised exchange in the outbound flow it traced, though it also notes that some proceeds are sitting in original or newly created wallets, so where all of it ends up is not settled.
What the report does and doesn’t establish
The 224 victims and 274.60 ETH cover one case: the nine videos TRM analysed. TRM says earlier versions have since been removed from YouTube and that the activity “extends beyond the nine videos included in this analysis”. The real totals are probably larger, and nobody has published a figure for them.
The description of this as a coordinated operation is TRM’s own inference, drawn from funds converging on six shared collection addresses and from stylistic similarities across the tutorials. It is not a formal attribution. No operator, group or individual has been identified, and this is a forensic analysis rather than a law enforcement action. The $517,205 figure is a valuation at the time each transfer happened, not a current or fixed loss.
TRM’s post also opens by citing its own AI-in-Crime Adoption Index, which it says shows criminal adoption of AI rising 40% year on year, and presents this case as an illustration of that trend. That framing belongs to the firm publishing it, and the case itself is bounded to the nine videos examined.
If this happened to you, and what to check before deploying anything
In England, Wales and Northern Ireland, fraud is reported to Action Fraud, online or on 0300 123 2040. In Scotland it goes to Police Scotland on 101. Recovery is unlikely once funds have crossed a bridge or a mixer, but reports are what build the case file that eventually names someone.
Two checks are worth knowing for anyone deploying a contract in future. Remix lives at remix.ethereum.org and nowhere else, so any compiler reached through a link in a video description deserves a hard look at the address bar. And the contract address produced by a deployment can be read on a block explorer before any money goes near it: if the code on-chain doesn’t match what was pasted in, something else was deployed. None of that is obvious, and none of it is taught alongside the advice to hold your own keys.
What to watch
Whether the remaining videos come down, and whether anyone publishes a count that covers the removed ones. TRM has effectively said the number it reported is a floor.
The six collection addresses are also public now. If those balances move toward an exchange with identity checks, this stops being a tracing exercise and starts being something a prosecutor can work with.