Seriously, Please Watch This Before AI Deepfakes Steal Your Crypto - Trezor CTO | E184

Tomas Sushanka, Trezor’s CTO, says even experienced security professionals can be fooled: he once fell for a fake shopping website. Fatigue and distraction can weaken judgment during ordinary tasks, while scammers may combine phishing emails, WhatsApp messages and phone calls, use fluent English, and build trust over weeks. AI makes voice impersonation easier; Sushanka says about two minutes of recorded conversation may be enough to recreate a voice. He avoids unknown-number calls, texts the caller, and calls back if the request seems relevant.

A deepfake call at a BTC Pro conference showed how a plausible identity can be used to deliver a dangerous request. Attackers impersonated a founder in a Microsoft Teams meeting and urged participants to install an extension; two employees refused and ended the call. Sushanka says unexpected downloads and installations should raise suspicion, even after several trust-building meetings. He recommends checking urgent requests through a separate channel and using browser-based calls where possible. Browsers are not risk-free, but persuading someone to install a malicious file is generally easier than exploiting a browser zero-day. AI can also help attackers write convincing messages and find malicious code, though it can help developers find software vulnerabilities.

Crypto recovery scams exploit people when they are distressed or distracted. One fake email targeting a former Celsius customer asked for a seed phrase, while a person who had lost hundreds of thousands of dollars later received fraudulent offers to recover money from a supposed $300,000 balance. Sushanka warns that entering recovery words on a phishing site can hand over funds. He says crypto’s risks are unfamiliar to many people, and convenience, limited education and overconfidence can undermine security even when the potential loss is large. Crypto attackers, he says, can be specialized and well funded; apparent business activity, including reported deposits of one or two million dollars by Drift Protocol attackers, can make a target less wary.

Self-custody means the user controls the Bitcoin, rather than relying on a provider’s database entry. It can offer security, availability and privacy, while custodial services may limit withdrawals or require personal documents; exchanges remain targets despite improvements since the Mt. Gox era. Sushanka says a hardware wallet may be worth considering when holdings approach the device’s price, or at roughly twice that value if the user intends to hold the crypto there. He favors moving funds gradually, starting with some and increasing as confidence grows. The right allocation depends on circumstances: he suggests 70–80% self-custody may suit some people, while estimating self-custody could suit about half of users. A small exchange balance for trading can be reasonable, but he sees little reason to leave an entire holding there.

Sushanka says self-custody is more manageable than it was a decade ago, provided users protect the device with basic measures such as safe storage and a PIN, and keep and verify their backups. Trezor’s recovery words generate private keys and are kept offline, rather than backed up to the cloud; he recognizes that this format can feel unfamiliar or intimidating. He says transparent, open-source tools allow users to access their coins through compatible software even if Trezor were to close. He sees inheritance as a difficult unresolved problem: a decentralized solution that avoids relying on companies such as Google or Apple is hard to create, while an ETF may simplify inheritance at the cost of dependence on a broker. He also notes that Trezor’s discontinued Model One lacks a secure element but will receive security updates for another 10 years.

On Trezor’s Safe 7, private keys are stored across three chips, including an Infineon secure element and a Tropic Square chip. Sushanka says there are no known remote exploits for the chips, but that does not prove none exist; an attacker would also need the device and would face protections including a PIN and potentially a passphrase. He does not expect a relevant quantum computer within five years and says it probably will not arrive within ten, but considers the risk low, not zero. Quantum computing could threaten cryptography used by Bitcoin and other systems, and he expects Bitcoin to adopt quantum-safe algorithms, though decentralized agreement on upgrades could take substantial time. Safe 7’s quantum-safe feature applies to firmware updates; it does not make Bitcoin itself quantum-proof.