TRM: fake Claude YouTube AI bot tutorials drained 274.60 ETH from 224 victims
By Crypto Wire
September 19, 2026
The newest wallet drain does not start with a phishing link. CoinGabbar wrapped a TRM Labs investigation on September 19 showing how fake YouTube tutorials that promise a Claude-built crypto arbitrage bot talk victims into deploying and funding their own malicious smart contracts. TRM’s primary write-up, dated September 14, traces 274.60 ETH—about $517,205 at transfer-time prices—from 224 victims into six operator addresses between February 12 and August 11, 2026.
Plain English Policy matters here because the attack path breaks the usual mental model collectors use. There is no spoofed marketplace, no malicious token approval prompt, and no “connect wallet” lure. Victims choose the tutorial, copy code, deploy a contract from their own wallet, and press the Start or Withdraw button the video tells them to press. Wallet phishing blocklists and approval-signature warnings never fire, TRM said, because every onchain action looks self-initiated.
TRM identified nine nearly identical tutorials across different channels that still looked independent on the surface. Scripts matched, claimed profits repeated down to the decimal—including a return of 1 ETH every 20 hours—and presenters or voices were AI-generated to fake variety. As of September 2026 those nine videos had about 310,474 views, with the earliest posted in April 2026. Earlier ChatGPT-branded versions of the same scheme appeared in 2025 before operators shifted the lure to Claude branding; Anthropic products never appear in the deployed contracts, TRM found.
The swap happens at the compiler. Tutorials link to operator-built sites styled like Remix, hosted on Amazon S3, Google Cloud Storage, or standalone domains. In one variant a background script discards the pasted source and compiles a different contract fetched from the operator’s server. The clean decoy code on screen never reaches the chain. The live bytecode is a honeypot that forwards any balance above 0.05 ETH to the operator when Start or Withdraw is called. Some sites then show a fabricated “gas nonce liquidity” error to squeeze a second deposit—TRM notes that phrase is not a real Ethereum concept.
Onchain, TRM mapped 234 victim-deployed contracts funding those six collection addresses. The median loss was 1 ETH, matching tutorial instructions to seed 1 to 2 ETH. Outbound flows stayed in DeFi swaps, bridges, and at least one mixer, with no centralized exchange in the path TRM published. For NFT freelancers and collectors who follow “build a bot / mint tooling” videos, the desk rule is simple: never deploy through a compiler a stranger linked in a description.
Attribute the September 19 circulation wrap to CoinGabbar’s TRM summary. Attribute the 274.60 ETH / $517,205 haul, 224 victims, 234 contracts, six operator addresses, February–August window, nine-video / 310,474-view tally, Remix-styled compiler swap, 0.05 ETH forward threshold, gas-nonce follow-up lure, Claude-vs-ChatGPT branding shift, and DeFi-only outbound path to TRM Labs’ September 14 investigation. Do not invent a named NFT collection drained or a YouTube channel list TRM did not publish in the summary figures above.
Plain-English takeaway: if a coding tutorial ships its own “Remix-like” deploy tool and an AI brand in the title, treat the compiler as the attack surface. The victim’s wallet will not warn them, because they are deploying and funding the contract themselves—exactly what TRM documented across this Claude-branded wave.
Collectors who hire contractors for mint sites or Discord bots should require verified toolchains—official Remix, Foundry, or Hardhat from known domains—and reject screen-share “interview compilers” or description links that load custom deploy UIs. That hygiene sits next to the WaterPlum hiring-malware alert already on the Rug Room desk: both campaigns abuse trust in professional workflows rather than classic phishing pop-ups.
TRM’s broader crime-index context—that criminal AI adoption rose about 40% year on year—belongs as backdrop, not as a separate victim count. The actionable number for this desk remains the Claude-tutorial cohort: 224 people, 274.60 ETH, and a compiler substitution that leaves the on-screen Solidity looking clean. Until YouTube removes the remaining nine videos, expect the same recording kit to keep rotating description links even after individual channels are reported.
Disclaimer: This article is provided for informational and educational purposes only. It does not constitute financial, investment, legal, or trading advice. The NFT market is highly volatile, and past performance is not indicative of future results. Readers should conduct their own research and consult qualified professionals before making any decisions related to digital assets. The cover image for this article may have been created using artificial intelligence (AI).
