The logic held; the trust was the vector.
A new phishing campaign has been dissected by SlowMist, and the pattern is textbook social engineering with a custom-built cross-platform malware suite. The attack vector is not a flash loan or a smart contract exploit—it is the human desire to secure a job in a shrinking market. The target: Web3 professionals. The bait: a fake AI meeting tool called 'Relay.' The result: your private keys, browser credentials, and Telegram sessions uploaded to a server you never authorized.
This is not a warning about a protocol bug. It is a warning about the fragility of trust in remote recruitment.
Context: The Hype Cycle of 'AI-Enhanced Hiring'
In 2025, the crypto industry is in a bear market. Hiring is down, but competition is up. Recruiters use AI tools to screen candidates, and candidates use AI tools to cheat on technical interviews. The ecosystem has normalized the installation of 'lightweight meeting software' for asynchronous interviews. Attackers exploit this normalization. They pose as recruiters from reputable firms, send a link to download 'Relay'—a custom-built info-stealer—and wait for the victim to enter their seed phrase or unlock their browser-stored credentials.
I have audited enough smart contracts to know that the weakest link is never the code—it is the person who runs untrusted binaries on their main machine. This attack is a textbook case of that principle.
Core: A Systematic Teardown of the Deception
The malware is not amateurish. SlowMist’s sample analysis reveals a multi-module stealer targeting both macOS and Windows. It scrapes browser databases for saved passwords and autofill data—a goldmine for accessing email accounts and exchange logins. It targets wallet extensions like MetaMask, Phantom, and Keplr by sifting through browser extension storage directories. It also extracts iCloud Keychain and local credential stores. But the most sophisticated touch is the Telegram session hijacker. By stealing the tdata folder, attackers can bypass two-factor authentication and impersonate the victim within their own professional DMs.
I traced the infection chain in my own sandbox. The installer calls home to a command-and-control server that appears to be hosted behind a commercial CDN, making takedown requests slower. The code does not lie, but it can be misled—here, the victim is misled by the professional facade of the recruiter’s profile.
The yield was not profit; it was a permission slip. The attacker’s goal was not to drain a single wallet. It was to harvest access credentials to infiltrate teams. Once inside a Telegram group for a DAO or a private Discord for a token launch, the attacker can rug-pull with insider information or social engineer other members.
Contrarian: What the Bulls Got Right
One could argue that this attack is a net positive for the ecosystem. It validates the value proposition of hardware wallets—a device that never exposes private keys to the operating system. It also reinforces the need for dedicated 'work devices' or virtual machine isolation for any interview process. The bullish take: this event will accelerate adoption of zero-trust architecture and decentralized identity (DID) solutions, where credentials are verifiable without running third-party binaries.
But the blind spot is that most Web3 workers still use their main machine for everything. The convenience of hot wallets and browser-based access trumps security until a breach occurs. The attack succeeds exactly because the ecosystem’s safety net—community vigilance, security checklists—is only as good as the last warning tweet.
Algorithmic fairness assumes fair inputs. Social engineering assumes honest actors. Neither assumption holds.
Takeaway: The Accountability Call
Every Web3 professional who reads this must perform a simple audit: have you installed any unsolicited meeting software in the last month? If so, disconnect that device from the internet, change all passwords from a clean device, rotate session tokens, and assume your Telegram account is compromised.
For the industry: this is not a one-off. Expect variants that use AI-generated video deepfakes of recruiters. The question is not whether your code is secure—it is whether your hiring process is secure. The next time a 'recruiter' sends you a link, ask yourself: how many wallets have drained because of a simple download?
Honest signals.
Your wallet doesn't scream when it's being emptied. The code does not cry when it's misused. But you can hear the silence of stolen funds.
Transparency is a feature, not a default state.
Until the hiring process forces verifiable, sandboxed interactions, this attack will be repeated with different names but the same payload.
Bots do not dream, they only scrape.
But the humans behind them dream of your seed phrase.
