A new attack vector has emerged that bypasses the usual smart contract exploit playbook. It doesn't target a vulnerability in Solidity or a flaw in a DeFi protocol. It targets the human layer, exploiting the very trust mechanisms that grease the wheels of Web3 hiring. SlowMist's latest disclosure reveals a sophisticated social engineering campaign where attackers pose as recruiters, luring Web3 professionals into installing a malicious 'AI meeting tool' called Relay. The payload? A cross-platform info-stealer that systematically extracts browser credentials, crypto wallet data, macOS Keychain secrets, and Telegram session tokens. It plays on the industry's current obsession with AI efficiency, but the code speaks louder than the whitepaper – and the code here is a weapon, not a tool.
Context: The Perfect Phishing Scenario The crypto bull market of 2025 has unleashed a frenzy of hiring. Projects flush with capital are competing for a limited pool of developers, auditors, and security engineers. This creates a fertile ground for social engineering. Attackers capitalize on the urgency of job seekers, the prevalence of remote-first interviews, and the uncritical embrace of AI tools. The fake 'Relay' app hides in plain sight, claiming to be an AI-powered meeting companion that transcribes conversations and summarizes action items. In reality, it is a custom-built trojan, compiled for both macOS and Windows, designed to plunder sensitive data from devices that hold the keys to millions in digital assets. SlowMist's analysis confirms the infection chain: a phishing email or LinkedIn message from a fake recruiter → a link to download 'Relay' → once executed, the malware installs a backdoor and begins exfiltrating data. The attack is not novel in technique, but its precision targeting of Web3 professionals makes it a critical threat.
Core: A Systematic Teardown of the Relay Trojan Let's dissect what the malware does, from a forensic perspective. The authors invested significant effort into cross-platform compatibility. For macOS, the compiled binary likely avoids Gatekeeper’s notarization requirement by exploiting a code-signing loophole or using a compromised developer certificate. I suspect it employs a combination of standard dylib injection and DYLD_INSERT_LIBRARIES to hook into browser processes. On Windows, it probably uses a similar DLL sideloading technique, coupled with persistence via registry keys or scheduled tasks. The trojan’s data collection scope is breathtaking: it iterates through Chrome, Firefox, Brave, and Opera profiles to grab saved passwords and cookies. It targets browser extensions for crypto wallets – MetaMask, Phantom, Keplr – stealing seed phrases and private keys stored unencrypted in extension local storage. On macOS, it extracts the user’s Keychain, which often contains API keys, exchange credentials, and SSH keys. On Windows, it scrapes the credential manager. The Telegram session hijacking is particularly insidious: it copies the tdata folder (or equivalent) to hijack active sessions, allowing the attacker to impersonate the victim within their work circles. This is not a script-kiddie tool. This is a purpose-built weapon for targeted destruction. Based on my audit experience, the code complexity suggests a team with prior malware development background, possibly APT-adjacent or ex-state actors repurposing techniques for cryptocurrency theft. The modular design – separate modules for each data source – indicates they plan to extend it. Future versions might grab Authenticator one-time codes or scan for SSH keys pointing to development servers. The attack chain is closed-loop: once the recruiter's impersonation is accepted, the victim's trust is the vulnerability that the malware exploits. As I often say, trust is a vulnerability vector. This campaign weaponizes the very premise of trust that underpins the hiring process.
Contrarian: What the Bulls Got Right Before the cynicism becomes too thick, let me address the counter-intuitive angle. The bulls – those who champion crypto adoption and dismiss security FUD as a distraction – have a point. Yes, this attack is serious, but it also reveals a healthy security ecosystem. SlowMist’s rapid disclosure and sample analysis are signs of a maturing industry that can detect and respond to threats quickly. The fact that the malware was discovered and publicly dissected means that intelligence-sharing mechanisms are working. Furthermore, the attack implicitly validates the importance of hardware wallets. If your Ledger or Trezor holds your assets, even a full compromise of your computer cannot extract the private keys. The bull case is that such attacks will accelerate the adoption of best practices: dedicated interview machines, multi-sig for corporate treasuries, and zero-trust architectures. The crooks are upping their game, but so are the defenders. The market may even see an increase in demand for security training and endpoint protection solutions tailored to Web3. The narrative of 'AI tools are dangerous' might actually reduce the hasty adoption of unvetted software, which is a net positive in the long term. However, this doesn’t excuse the underlying complacency. The flaw in the industry’s approach is that it prioritizes on-chain security while ignoring off-chain attack surfaces. Smart contract audits are necessary but insufficient when the user’s laptop is the weakest link. Complexity is the enemy of security. The industry has added layers of infrastructure – wallet extensions, bridges, liquid staking – but has failed to simplify the user’s operational security.
Takeaway: The Human Layer Is Now the Primary Attack Surface This incident is a harbinger. The next wave of crypto theft will not come from a reentrancy bug or an oracle manipulation. It will come from compromised keyboards, stolen browser histories, and hijacked Telegram sessions. The code speaks louder than the whitepaper, but in this case, the code is the social engineering script. Every artifact is a trace of failure. The failure here is not just of the victims, but of an industry that hasn’t institutionalized basic hygiene for its most valuable participants. Logic does not bleed, but it does break. And when trust becomes the attack vector, the only defense is a paranoid, adversarial mindset. Audit first, trust never – but that maxim must extend beyond smart contracts to every link in the chain. If you receive an unsolicited interview invitation, assume breach. Verify the source through multiple channels. Use a dedicated, isolated machine for interviews. And above all, store your private keys on a metal box, not a corner of your file system. The market may be euphoric, but the grim reaper is phishing for keys. Act accordingly.