They came as recruiters. They offered AI interview tools. They took everything.
On July 29, 2025, SlowMist released a chilling analysis. A social engineering campaign, targeting Web3 professionals, is live. The lure: a fake AI meeting app called "Relay." The payload: a cross-platform info-stealer that drains crypto wallets, hijacks browser credentials, and exfiltrates Telegram sessions. The real target: your private keys.
I have seen this pattern before. In 2017, the Parity heist exploited a reentrancy bug in a wallet library. In 2020, the Curve treasury drain revealed the danger of compromised hot wallets. But this attack is different. It bypasses code audits. It exploits trust, not technology. And it is happening right now.
Context: The Weaponization of AI Hype
The crypto job market is booming. Bull market euphoria has companies scrambling for talent. Recruiters flood LinkedIn, Telegram, Discord. AI tools are the new normal—automated screening, AI interview bots, virtual meeting assistants. The attacker weaponized this narrative.
They pose as hiring managers from legitimate protocols—or fake ones—and invite targets to a "technical interview" using Relay, an AI-powered meeting scheduler. The victim downloads a desktop app. For macOS, it's a signed DMG. For Windows, an executable. Both install silently. Both steal.
This is not a phishing link. This is a full-scale malware deployment disguised as productivity software.
Core: The Technical Anatomy of the Theft
Based on the sample analysis published by SlowMist, the malware performs the following:
- Browser credential theft: It targets Chrome, Brave, Edge, and other Chromium-based browsers. It extracts saved passwords, cookies, and autofill data. For crypto professionals, that often includes exchange logins and 2FA backup codes.
- Crypto wallet extraction: Specifically, it accesses browser extension wallets—MetaMask, Phantom, Keplr, and others. It reads the
Local StorageandIndexedDBdatabases where encrypted private keys reside. If the wallet is unlocked (which it often is during an active session), the keys are decrypted locally and exfiltrated. - Keychain/iCloud Keychain dump: On macOS, it attempts to dump the system keychain. On Windows, it queries the Credential Manager. This yields API keys, VPN credentials, and potentially SSH keys.
- Telegram session hijack: It reads Telegram Desktop’s local session files (
tdata) and exfiltrates them. This allows the attacker to impersonate the victim within their network—sending malicious messages to colleagues, pretending to be the user, and spreading the attack laterally.
The exfiltration is likely via encrypted C2 channels. SlowMist identified specific IPs and domains—I strongly urge readers to check their security feeds for the full IOC list.
But what is the attack’s true innovation? It’s the timing and targeting. The malware only activates during the supposed "meeting" window. It uses the target’s microphone and camera as a distraction—while the victim believes they are speaking to an AI bot, the malware executes silently in the background. The AI bot is likely just a recorded video or a simple script. The real recruiter is not real at all.
Contrarian Angle: The Attack Vector Nobody Is Talking About
SlowMist nailed the technical details. But the industry is focusing on the wrong fix. Everyone says: "Don't install untrusted apps." That’s useless advice in a bull market where new tools appear daily.
The real blind spot is identity verification on Web3 hiring platforms. LinkedIn, Telegram, and Discord have no standardized way to prove a recruiter is legitimate. Attackers create fake profiles with stolen credentials, sometimes even hijacking real accounts. The victim’s first line of defense—the platform—is already compromised.
I have tracked this problem since 2021, when I was excluded from the Bored Ape YCIP-001 drafting because internal politics prioritized hype over process. The same lack of provenance haunts recruitment today. A legitimate protocol should provide a verifiable on-chain credential for its hiring team. A simple signed message from a corporate multisig could eliminate this entire class of attack.
Furthermore, the malware’s ability to steal Telegram sessions is a second-order threat. Attackers can now impersonate the victim within their professional network, sending fake meeting invites to colleagues. This creates a cascading breach. One compromised developer can lead to a full protocol treasury drain.
Speed is safety when the exploit is already live. But prevention requires structural changes to how we verify identities in Web3.
Takeaway: What to Do Now—and What to Watch For
- Immediate actions: If you have received any unsolicited interview invitation that asks you to install a meeting app named Relay or any unknown software, freeze. Do not open it. Check the domain of the recruiter’s email against the official company domain. Use a dedicated virtual machine or live USB for any untrusted application.
- Long-term fix: Demand that protocols implement a public key infrastructure for hiring. Corporate entities should sign their job postings with a recognized on-chain identity. If you are a hiring manager, adopt web3-native verification—use ENS or a verified Coinbase account, not a LinkedIn link.
- Next watch: SlowMist will likely release updated IOCs. Track the C2 servers. If the attack expands to deepfake video interviews, the damage will multiply. The chart doesn’t lie, but the recruiter’s profile might.
We don’t have the luxury of waiting for the next exploit. The malware is already in the wild. Trust your code review instincts—apply them to your hiring process.