We teach code as law, yet the most devastating exploits rarely target a smart contract. They target the human behind the screen. On July 29, 2025, SlowMist disclosed an attack chain that perfectly illustrates this paradox: a fake AI-powered interview tool called "Relay" that steals cryptocurrency wallet data, browser credentials, and Telegram sessions from Web3 professionals. The malware runs on both macOS and Windows. The psychological trap is simple—promise a job at a crypto startup via LinkedIn, ask the candidate to install a "meeting scheduler," and drain everything before the interview even begins.
Allow me to step back and frame why this matters beyond the technical IoCs. Over the past eight years, I've watched the industry oscillate between euphoria and panic, each cycle teaching us that the most resilient protocols are those that anticipate human fallibility. I cut my teeth auditing early ERC-20 distribution logic in 2017, where the vulnerability wasn't in the code but in the power asymmetry between whales and retail holders. I later spent the 2020 DeFi Summer running community education circles at Aave, watching newcomers lose savings not because of a bug, but because they trusted a flashy yield dashboard. And in 2022, during the governance crisis at Compound, I mediated forums where developers and users alike realized that trust destroyed by a single phishing email takes months to rebuild. Each experience confirmed a principle I now hold as foundational: "Resilience beats hype every time."
Now, in mid-2025, the market is treading sideways. Traders are hungry for direction signals, but the real signal is not a breakout pattern—it's the quiet evolution of threat vectors. The "Relay" attack is a canary in the coal mine for how decentralized talent markets become centralized honeypots. Let me unpack the technical mechanics, the stewardship failure it reveals, and why the contrarian takeaway is that we need less trust in endpoints and more trust in communities.
The Tech Behind the Trap
The attack chain is deceptively elegant. The adversary creates a convincing LinkedIn profile posing as a recruiter for a known Web3 protocol. They message a target—typically a developer, security researcher, or community manager—with a "pre-interview test" or "AI-based scheduling tool" called Relay. The target downloads an executable that masquerades as a legitimate video meeting client but is, in fact, an information stealer compiled for both macOS and Windows. According to SlowMist's sample analysis, once executed, the malware extracts:
- Browser-stored passwords and cookies (Chrome, Brave, Firefox)
- Crypto wallet extensions (MetaMask, Phantom, Keplr, etc.)—including seed phrases if stored unencrypted
- macOS Keychain data
- Telegram session cookies
- SSH keys and configuration files
What distinguishes this from generic infostealers is its precision. It targets credentials that unlock not just a bank account, but a professional's entire crypto ecosystem: Telegram groups for DAO coordination, multisig signer sessions, and even personal wallets holding governance tokens. The cross-platform capability indicates a developer with deep knowledge of both operating systems' security boundaries. Based on my experience auditing smart contract security, I've seen similar sophistication only in advanced persistent threat (APT) groups—though here the aim is financial gain, not espionage.
But the most striking aspect is the social engineering payload. The name "Relay" capitalizes on the growing trend of AI-enhanced recruitment tools—"AI interviews" and "automated scheduling" have become buzzwords. The attacker understood that Web3 professionals are accustomed to using new, unvetted tools in the name of agility. They leveraged the very culture of "move fast and experiment" that built the ecosystem.
Code Is Law, But People Are Purpose
This attack forces us to confront a stewardship failure. We have spent years optimizing for code security: formal verification, zk proofs, multi-signature schemes, and secure enclaves. Yet we have neglected the human interface layer. Every hardware wallet, every audited smart contract, every zk rollup becomes irrelevant the moment a user voluntarily hands over their private keys to a fake installer. "Trust, verify. But also, connect." The verification must extend beyond cryptographic signatures to the social relationships we embed trust in.
Here I must invoke my own scars. During the 2022 bear market, I managed the Compound community's transition through a governance crisis. The emotional toll was immense—hundreds of users messaging me in despair after losing funds to phishing attacks that exploited trust in familiar names. I realized then that our industry's resilience is not a function of gas optimizations or TVL rankings. It's a function of how well we inoculate our communities against social engineering. "Community is the new central bank." Not because it holds reserves, but because it validates identity and intent.
The "Relay" attack directly exploits the gap between technical sophistication and social naivety. The victim trusts the recruiter's profile (which may have been hacked or fabricated with AI-generated images), the familiar-to-use tool (Relay sounds like Zoom or Calendly), and the promise of career progression. No amount of protocol-level security can stop that if the user runs the wrong binary.
The Contrarian Angle: Hardware Wallets Are Not Enough
The immediate industry reflex will be to recommend hardware wallets—and yes, they mitigate asset theft from non-custodial wallets. But that advice is dangerously incomplete. The attacker also steals browser credentials that grant access to centralized exchange accounts, Telegram sessions that allow impersonation of the victim to drain their social capital, and SSH keys that could lead to corporate infrastructure compromise. A hardware wallet does nothing against a stolen API key for a DeFi lending position.
Moreover, the attack vector itself evolves. If the community over-indexes on hardware wallet adoption, attackers shift to exploiting hot wallet browser extensions that bypass cold storage. Or they pivot to sim-swapping recovery methods. The real blind spot is our assumption that endpoint security is a solved problem because we run antivirus or use Passkeys. On macOS, Apple's Gatekeeper can be bypassed by signing the malware with a stolen developer certificate—which is exactly what sophisticated stealer campaigns do. Windows Defender similarly fails against novel, custom-compiled binaries.
So what is the pragmatic mitigation? I advocate for what I call "zero-trust interviewing": treat each job application as a secure enclave session. Use a dedicated, ephemeral virtual machine that is wiped after each interview. Never sign into personal Telegram, email, or wallet on the same machine used for unknown executables. And verify the recruiter through a secondary channel—call the company's official phone number or ask for a one-time code sent to an independently obtained email.
But these are individual tactical fixes. The deeper strategic failure is structural: the Web3 hiring ecosystem lacks decentralized identity primitives. Many protocols claim to use decentralized identities (DIDs) or verifiable credentials, but few job applicants actually require a recruiter to prove their affiliation through a signed message from the protocol's multisig or governance forum. We need to normalize "verify before install" as a community norm. "Trust, but verify. But also, connect." The verification must be built into the workflow, not left as user responsibility.
The Forward-Looking Takeaway
The "Relay" malware is not an anomaly; it's a prototype for a class of attacks I've been tracking since 2023. As AI-generated faces and voices become indistinguishable from real ones, social engineering will shift from text-based phishing to full video impersonation. Imagine a Deepfake version of a protocol's co-founder asking you to install an urgent update during a fake virtual all-hands. Our current defense—user education—will fail against attackers who can mimic a trusted colleague's voice with 99% accuracy.
The only sustainable solution is to create protocols that assume the endpoint is compromised and still preserve integrity. That means moving seed phrases off main memory (use hardware signatures), requiring out-of-band confirmation for any transaction or credential grant, and embedding social recovery mechanisms that do not rely on a single password or session.
I end where I began: "Code is law, but people are purpose." The purpose of our technology is to empower human agency, not to expose it to new vulnerability. If we build systems that are resilient to human error—rather than systems that punish it—we will survive every bear market and every attack wave. Resilience beats hype every time.
Listen to the signal this attack sends: the blockchain is secure; the human chain is not. Strengthen the second one, and the first will take care of itself.