The 'Relay' Trap: How a Fake AI Interview Tool Is Draining Web3 Wallets

You open your laptop, coffee in hand, ready for a remote interview with a top Web3 venture fund. The recruiter on LinkedIn was polished, the role perfect for your skills. They send you a link to 'Relay,' an AI-powered meeting platform promising seamless performance. You download, install, and give it screen recording permissions. Two hours later, you notice your Ledger app won't sync. Your hot wallet shows zero balance. A Telegram session you never started is sending messages to your contacts. You've been hit. Not by a phishing email, but by a cross-platform data-scraper disguised as a hiring tool. This isn't a hypothetical. On July 29, 2025, SlowMist Security released a detailed analysis of a new attack wave targeting Web3 professionals exactly like you. The attackers pose as recruiters from legitimate crypto funds or exchanges, using fake LinkedIn profiles and convincing job descriptions. They ask candidates to install 'Relay,' a custom-built malware that masquerades as an AI meeting assistant. Once installed, it steals browser credentials, cryptocurrency wallet files, macOS Keychain entries, and Telegram session data. The malware is compiled for both macOS and Windows, demonstrating a sophisticated, platform-agnostic approach. The narrative hook — 'AI-powered interview tool' — leverages the current hype around AI agents and remote work, making it almost irresistible to eager job seekers. Let's rewind. Web3 has always been a target-rich environment for social engineers. From the early days of Mt. Gox phishing to the recent spate of fake NFT mints, the industry runs on trust and speed. But this attack is different. It's not a mass email blast; it's a precise surgical strike. The attackers spent time building credible identities, researching their targets (probably on LinkedIn, Discord, and Telegram), and crafting a custom payload that evades basic antivirus scans. The 'Relay' malware is no script-kiddie tool. Based on my own work reverse-engineering malicious samples post-Terra — when I spent three months auditing Arbitrum's fraud proofs — I recognize the hallmarks of a team with both motivation and deep technical skill. The multi-platform nature alone suggests a developer familiar with Windows PE and macOS Mach-O structures. And the theft of Telegram sessions is a clever multiplier: once inside a victim's chat, they can impersonate the user to reach other high-value targets within the same network. The core of this attack lies in its exploitation of two cognitive biases: authority (the fake recruiter) and utility (the promise of an AI tool). The malware likely uses code obfuscation and possibly signed binaries to bypass Gatekeeper on macOS or Windows Defender. After installation, it quietly scrapes a predefined list of directories and registry keys, then exfiltrates the data over HTTPS to a command-and-control server. SlowMist's report includes specific hashes and indicators of compromise (IOCs), but the real signal is the pattern: attackers are weaponizing the very narratives that drive our industry forward — AI, remote work, and meritocratic hiring. Mapping the chaos to find the signal in the noise, I see a clear trend: the intersection of AI hype and social engineering is the new frontier of crypto crime. Now for the contrarian angle. Most security reports will scream panic, urging you to never install any third-party software. But that advice is both impractical and naive. The real lesson is about verifiable identity and isolation. In Web3, we preach 'don't trust, verify' for smart contracts, yet we often ignore it for human interactions. The contrarian take is that this attack, while tragic for victims, will ultimately strengthen the ecosystem. It forces our community to embrace better practices: using hardware wallets as the sole signing device, running all untrusted applications in ephemeral virtual machines, and demanding decentralized identity verification (DID) for any financial interaction. Stories drive value, not just algorithms — and the story of 'Relay' will push hiring platforms like LinkedIn to integrate on-chain verification or risk losing credibility. From the ashes of Terra, we learned to walk; from these phishing ashes, we learn to verify. This is evolution through pain, and it's the only path to resilience. Where does this leave us? The 'Relay' scam is a trial run. The next iteration will involve deepfake video calls, AI-generated voice cloning, and fully automated recruiter bots. We're not just fighting malware; we're fighting narrative engineering. My takeaway is this: treat every job interview as a zero-trust exercise. Demand a video call that you initiate, never click a direct download link, and use a dedicated machine or sandbox for any new software. The future of Web3 security isn't just in code audits — it's in institutionalizing skepticism in our daily workflows. Rebuilding the compass after the storm passes means integrating threat intelligence into our own patterns. The attack on 'Relay' is a spark. But if we use it to light a fire under our own security practices, we turn a crisis into a catalyst. The question is: will you change your habits today, or wait until your own wallet is drained? Hunting for the next spark in the dry brush — the next attack is already being planned. Let's make sure we see the smoke before the fire.