Imagine this: you are a blockchain developer, actively searching for your next role. A LinkedIn recruiter messages you with an enticing offer—a paid trial using an innovative AI meeting tool called Relay. You download it, install it, run it. The next day, your hot wallet is empty. Your Telegram account is compromised. Your browser credentials are leaked. This is not a hypothetical. This is the new attack vector revealed by SlowMist on July 29, 2025.
Logic does not bleed, but code leaves traces. And the traces here are alarming.
Context: The Attack Surface of Trust
The crypto industry runs on trust—trust in code, trust in communities, trust in job offers. Attackers know this. Over the past months, a new wave of social engineering has emerged, targeting exactly this vulnerability. The attack begins with a phishing message on LinkedIn or similar platforms. The recruiter impersonates a legitimate crypto firm, inviting the target to interview using an AI meeting tool called Relay. The tool is not real. It is a custom-built infostealer, compiled for both macOS and Windows, designed to exfiltrate sensitive data with surgical precision.
Based on my experience auditing DeFi exploits and tracking wallet drainers, I have seen social engineering evolve from simple phishing links to sophisticated multi-stage attacks. This one stands out because of its specificity: it targets Web3 professionals—people who hold significant crypto assets and have access to private keys, seed phrases, and exchange credentials. The attack chain is complete: lure, payload, exfiltration, monetization.
SlowMist’s public disclosure includes sample analysis, IOCs, and attack flow details. This is not just a warning; it is a forensic reconstruction of a real-time threat.
Core: The Machinery of Deception
Let me dissect the technical anatomy of this attack. The executable—named something innocuous like “Relay.dmg” or “RelayInstaller.exe”—is a trojanized application. On macOS, it likely uses Apple’s Developer ID signing to bypass Gatekeeper, or it may be delivered as an unsigned app that users are socially engineered to run via “right-click open.” On Windows, it probably exploits trusted publisher certificates or uses a standard .exe that triggers User Account Control but is granted permission by the user.
Once installed, the malware performs several actions:
- Browser Credential Theft: It targets stored passwords, cookies, and autofill data from Chrome, Firefox, Brave, and other Chromium-based browsers. This gives attackers access to exchange accounts, email, and social media.
- Crypto Wallet Extraction: It scans for common wallet extension directories (e.g., MetaMask, Phantom, Keplr, Rabby) and attempts to read the native encrypted keystore or even memory dumps if the wallet is unlocked. On macOS, it can access the Keychain, which often contains private keys if users saved them.
- Telegram Session Hijacking: The malware targets the Telegram desktop client’s session files. By copying the tdata folder (Windows) or the corresponding macOS directory, the attacker can impersonate the victim without needing 2FA codes. This is especially dangerous because Telegram is the primary communication channel for many crypto teams.
- Persistence: It installs login items (macOS LaunchAgents) or scheduled tasks (Windows Task Scheduler) to survive reboots and ensure continued access.
This is not a simple keylogger. It is a modular infostealer with a focus on crypto assets. The code is likely obfuscated—based on the complexity of the 10+ data exfiltration endpoints, the creators have invested significant resources. They understand the tools used by Web3 professionals.
The critical insight: This attack exploits the trust placed in the hiring process. The victim assumes the application is legitimate because the recruiter appears credible. But the rug is not pulled; it was never tied. The recruitment narrative is a veneer for a targeted infiltration.
What makes this particularly dangerous is the cross-platform implementation. Many crypto users on macOS believe they are immune to malware. This is a dangerous myth. The macOS version is just as capable as the Windows variant, leveraging Swift or Objective-C runtime to access the Keychain and file system.
From my analysis of similar attack patterns in 2023-2024 (e.g., the Cthulhu Stealer on macOS, the Mars Stealer on Windows), the evolution is clear: attackers are now tailoring malware to specific professions rather than casting wide nets. This attack is not a spam blast; it is a spear-phishing operation using AI-themed lures. The potential victim pool is smaller but far more lucrative.
Risk assessment: High. The success rate for this attack depends on the victim’s security hygiene. If the user stores private keys in browser extensions or saves passwords in the browser, the malware strikes gold. If they use a hardware wallet and never install untrusted software, the risk is mitigated. But the attack preys on the human factor—the desire to get a job, the curiosity to try a new AI tool.
This attack also reveals a supply chain vulnerability in recruitment. Web3 companies often use third-party talent acquisition platforms and contractors. Attackers can hijack legitimate recruiter accounts or create convincing fake profiles. The identity verification processes on LinkedIn are insufficient to prevent this.
SlowMist has shared the hash values, C2 domains, and file paths. But by the time you read this, the attackers may have updated the malware binaries. Static IOCs have a short shelf life. The real defense is behavioral: never install software that is not explicitly required by the company’s official IT onboarding process. If a recruiter asks you to install a new tool for an interview, it should be a red flag.
Contrarian: What the Bulls Get Right
Now, let me force myself to consider the opposing view. Some optimists might argue that this attack is not as widespread as it seems, that SlowMist’s disclosure is protective, and that the industry is already aware of such threats. They might point out that the attack only works if the user is foolish enough to install unknown software. In a sense, they are right—this is not a protocol exploit; it is a user error exploit.
But this perspective misses the bigger picture. The contrarian truth is that attacks like these are a sign of a maturing industry. When the value at stake is high enough, attackers invest in custom tools. This attack validates that Web3 is no longer a fringe experiment—it is a target-rich environment. The positive angle: such incidents force security improvements. Within six months, we will likely see mandatory security onboarding checklists for crypto companies, or even dedicated “interview machines” using virtual machines.
Another contrarian take: The AI narrative here is a double-edged sword. The attackers used the AI buzzword to lower defenses. But in doing so, they also highlighted a growing attack surface: AI-powered recruitment tools. This could accelerate adoption of zero-trust architectures and decentralized identity systems (e.g., using ENS domains or DIDs to verify recruiter authenticity). The attack, while damaging, might ultimately lead to a more resilient ecosystem.
Volume is noise; the wallet cluster is signal. The signal here is clear: social engineering is the most effective vector in 2025. No exploit is needed when you can trick a user into giving you keys.
Takeaway: Accountability and Forward-Looking Judgment
The question is not whether you will receive such a message; it is whether you will be ready when you do. Gas fees are the price of truth. The truth is that this attack will be refined and repeated. The next variant might use deepfake audio or video. The recruiter might be a fully AI-generated persona. The signature I want to leave you with: Imagination is infinite, but liquidity is finite. The attackers have infinite imagination; your liquidity is finite. Protect it.
Check the contract, not the influencer—but in this case, check the recruiter, not the tool. Use hardware wallets. Isolate your work environment. Assume every unsolicited software installation request is a trap. SlowMist did its job. Now it is your turn.
The rug is not pulled; it was never tied. But your security—that is something you can tie down today.
