A senior DeFi developer at a prominent Layer-2 protocol receives a LinkedIn message. The sender profile shows a polished headshot, a history of posts praising the latest zk-rollup, and a job title that reads “Talent Partner” at a well-known venture-backed project. The job offer is too specific—tailored to the developer’s GitHub history and conference talks. The recruiter invites them to an “AI-powered interview” using a tool called Relay. The developer, eager and trusting, downloads the installer. Within minutes, the damage is irreversible.
This is not a hypothetical scenario. On July 29, 2025, SlowMist disclosed a real-world attack chain where threat actors impersonate Web3 recruiters, distribute a custom information-stealer disguised as an AI meeting application, and exfiltrate browser credentials, crypto wallet data, keychain contents, Telegram sessions, and other sensitive assets. The malware targets both macOS and Windows, a rare cross-platform capability that reveals a sophisticated, well-resourced adversary. For anyone in the Web3 workforce, this is not a mere reminder to be careful—it is a structural failure in the trust layer that underpins decentralized hiring.
Context: The Decentralized Workforce’s Open Wound
The Web3 industry has long prided itself on borderless, permissionless collaboration. Remote hiring is the norm. DAOs, foundations, and startups rely on platforms like LinkedIn, Telegram, and Discord to source talent. Trust is commodity: a developer’s reputation is their code, their contributions, and their word. This openness, however, creates a vector that traditional social engineering attacks have increasingly exploited. In 2024 alone, we saw a 340% rise in fake job offers targeting crypto professionals, according to Chainalysis. But this Attack is different.
Past incidents were often crude—fake PDFs with macros, phishing links, or simple credential harvesting. This attack is surgical. The malware is not a repurposed commodity crimeware; it is purpose-built for the Web3 demographic. The attackers understood that a Web3 professional’s laptop holds more than just passwords—it holds their sovereign identity. Private keys, browser wallets, encrypted messaging sessions that grant access to DAO treasuries, Telegram groups where alpha is shared, and even SSH keys for cloud infrastructure. Silence in the chain speaks louder than noise—and here, the silence was the weeks of reconnaissance before the trigger.
I remember auditing a vesting contract in 2017 for a Lagos-based fintech. A colleague was nearly tricked by a phishing email that looked exactly like a company memo. That experience taught me that trust is a protocol, not a promise. In Web3, we build protocols for transactions, but we have not built protocols for professional identity verification. This attack exploits that gap.
Core: The Anatomy of a Targeted Credential Heist
Let us deconstruct the technical architecture of this threat, based on SlowMist’s analysis, public IOC data, and my own experience auditing code that handles sensitive data. The attack is a multi-stage social engineering plus custom malware operation.
Stage 1 – Reconnaissance and Impersonation The attackers create convincing recruiter profiles on LinkedIn and other platforms. They use AI-generated headshots, fabricated employment histories at real Web3 companies, and even interact with posts to build credibility. The target is a user with public GitHub, published research, or conference presentations—any signal that suggests they have valuable access. The recruiter messages them with a role that matches their expertise, often in DeFi, security, or Layer-2 engineering.
Stage 2 – The “Relay” Hook The recruiter pitches an “AI-powered interview tool” that supposedly integrates real-time code assessment, live collaboration, and automatic transcription. The name “Relay” evokes efficient communication. The victim is directed to a landing page that mimics a legitimate software site, complete with feature lists, testimonials, and download buttons for macOS and Windows. The installer is signed with a reputable-seeming certificate (likely stolen or fraudulently obtained). Once executed, the malware extracts itself from a packed archive and proceeds.
Stage 3 – Data Exfiltration The malware performs a systematic sweep of the victim’s machine. From the SlowMist report, the following data categories are targeted:
- Browser credentials (cookies, saved passwords, autofill data) from Chrome, Firefox, Brave, and other Chromium-based browsers.
- Crypto wallet data – extension wallets like MetaMask, Phantom, Keplr, and their associated seed phrases or encrypted vaults. The malware specifically looks for common wallet directories and configuration files.
- macOS Keychain – on macOS, it dumps the system keychain, which stores everything from Wi-Fi passwords to private keys for software wallets.
- Telegram session data – it steals the local Telegram Desktop database, allowing the attacker to hijack active sessions, read messages, and impersonate the victim in group chats.
- File system – it searches for common patterns like “.txt”, “.doc”, “.xlsx” containing keywords such as “seed”, “backup”, “mnemonic”, “keystore”.
The exfiltration is done over HTTPS to a command-and-control server that appears to be a legitimate cloud service. The malware also tries to disable or evade common endpoint detection by waiting for idle CPU time and avoiding sandboxed environments.
Why this attack works It works because it weaponizes the very trust that makes Web3 inviting. A developer, already conditioned to trust a recruiter with a real company logo, sees an AI tool as plausible innovation—not a threat. The cross-platform nature ensures no operating system is a safe harbor. The attackers have invested in signature evasion, signing, and infrastructure. This is not script kiddie work. It is a professional operation, likely with ties to a cybercrime group that specializes in crypto theft.
I have seen similar patterns in my own work auditing DAO governance proposals. One proposal was a stealth attempt to insert a backdoor into a treasury delegate system, wrapped in a “security enhancement” narrative. The authors had studied the community's technical language. Culture compiles where logic fails—human bias and desire for career advancement compile over cold analysis. The Relay malware is an extension of that principle.
Contrarian Angle: The Real Vulnerability Is Not the Malware
Conventional wisdom says: update your antivirus, don’t download unknown software, use hardware wallets. All correct. But the deeper issue is structural. The Web3 hiring pipeline is built on centralized identity platforms (LinkedIn, Telegram) that lack cryptographic guarantees. We preach decentralization for finance, but we rely on opaque databases for professional trust. The contrarian truth is that this attack is a symptom of a misaligned infrastructure.
Consider: if the job offer had been verified via a decentralized identity (DID) proof—where the recruiter’s DAO or company signed a verifiable credential attested on-chain—the victim would have a cryptographic basis for trust. Instead, we use profile pictures and mutual connections. The attack also reveals a blind spot in the “code is law” narrative. Code cannot enforce human interaction patterns. Social engineering exploits the gaps between blocks—the gray areas where protocol definitions end and human judgment begins. We govern these gray areas poorly.
Moreover, the financial incentive for attackers is enormous. A single compromised Telegram session can lead to a treasury drain worth millions. The cost of developing a cross-platform info-stealer is trivial compared to the potential take. This is asymmetric warfare. Until the Web3 industry collectively adopts a zero-trust identity layer for professional interactions, attacks like these will remain profitable. The contrarian view is that no amount of user education will solve this—technology must fill the trust vacuum.
Takeaway: Building Cathedrals in the Bear Market of Trust
We are in a bull market for scams. The euphoria and urgency of hiring create perfect conditions for predators. The Relay malware is a wake-up call, not for individual caution, but for systemic re-architecting. DAOs and Web3 companies must treat professional identity verification with the same rigor as smart contract audits. Every job offer should be verifiable via a signed message from a known entity. Every interview tool should be sandboxed or run in an ephemeral virtual machine. Every candidate should assume that unsolicited software is a threat until proven otherwise.
I propose a practical step: create a “safe interview protocol” on-chain. A recruiter publishes a commitment to their job listing, signed by a key associated with the organization’s multisig. The candidate verifies this commitment before downloading any software. This is not radical; it is a logical extension of the trustless paradigm that blockchains enable. Vision without verification is just hallucination—we must embed verification into the hiring process itself.
As for the immediate defence: use hardware wallets for all crypto interactions, keep them disconnected during interviews. Run interview tools in a dedicated virtual machine that has no access to your private keys or Telegram session. Change your Telegram password and enable two-factor authentication immediately. And remember, tokens are the brush, community is the canvas—but if the artist’s hand is compromised, the canvas is empty.
The battle against social engineering is not new. It predates crypto. But in Web3, the stakes are higher because the capital is directly at the edge of the user’s device. We must accept that intuition audits the code before the compiler does—and that intuition must be hardened by protocol. The Relay attack is just one brushstroke; a more sophisticated version will come. We have a narrow window to build the trust infrastructure that can prevent the next generation of these targeted heists. I am writing this as a practitioner who has seen what happens when code is correct but human trust is exploited. Trust is a protocol. It is time to compile it.