Former Binance Chief Executive Changpeng Zhao, who founded the exchange in 2017 and stepped down from the role in 2023, has pushed back against warnings that artificial intelligence will upend cryptocurrency security, arguing that fears of AI-driven hacking in the sector are overstated and that the internet and traditional banking already run on weaker cryptographic foundations than blockchain networks.

The remarks place the executive, widely known as CZ, at odds with a growing chorus of voices warning that advancing AI capabilities could eventually threaten the cryptographic primitives underpinning Bitcoin and other proof-of-work and proof-of-stake networks. In his view, the threat is real in theory but is frequently misframed when applied specifically to crypto, given that more vulnerable legacy infrastructure has not collapsed under comparable AI-assisted attack pressure. The debate has taken on added urgency as AI agents move from research demonstrations toward systems capable of interacting with live financial infrastructure.

The Comparative Case That Crypto Is Not the Weak Point

CZ's core argument rests on a comparative observation. The internet's foundational transport-layer security and the authentication systems used by banks rely on cryptographic standards that are, by several technical measures, less robust than the elliptic-curve and hash-based cryptography securing Bitcoin wallets and smart contract execution environments. If AI-powered attacks were ever to systematically break public-key cryptography, those older and more widely deployed systems would fall first and fastest.

That framing shifts the debate away from whether AI poses a cryptographic risk in the abstract and toward the question of relative exposure. Critics of crypto often point to AI-accelerated brute-force attacks, large language model-assisted phishing aimed at seed phrases, or AI agents exploiting flaws in smart contract logic. Under CZ's comparison, the first category is not a crypto-specific problem, and framing it as one overstates the sector's unique vulnerability.

The intersection of AI and crypto infrastructure is already operational rather than purely hypothetical. OpenAI has acknowledged an incident in which an AI agent accessed Ethereum testnet wallet funds.

The distinction matters because it separates catastrophic cryptographic-break scenarios from the more realistic near-term threat vector: social engineering. AI-generated deepfakes, voice cloning, and highly personalized phishing attacks do not require breaking elliptic-curve cryptography. They exploit human factors at the wallet and exchange interface layer, a risk that applies equally to banking customers and crypto holders.

What the Banking Comparison Actually Implies

CZ's comparison carries a secondary implication that cuts against the pessimistic view: systemic risk in crypto is not isolated. Legacy financial rails, running on protocols built decades before modern adversarial AI existed, represent a larger and softer attack surface when measured by total assets under management and user count. A sophisticated AI capable of cracking crypto-grade cryptography would render HTTPS, bank login authentication, and interbank settlement infrastructure equally obsolete.

That is not the same as saying AI poses no security risk to crypto. The argument is one of prioritization and proportion. The claim that AI will specifically and disproportionately devastate crypto while leaving the broader internet economy intact is difficult to sustain technically. Read carefully, CZ's position is a proportionality argument rather than a dismissal of security as a concern.

The stakes of the debate are nonetheless perceived as high. Investor Chamath Palihapitiya has separately argued that a successful hack of Bitcoin could send its price to zero, a scenario that underscores how consequential the security question is viewed to be, even as experts disagree on its probability.

The more operationally grounded near-term risk is AI-assisted social engineering at scale, not cryptographic collapse. Large language models can now generate highly convincing impersonation messages, synthesize the voice audio of known individuals, and identify high-value targets from public blockchain data with far less manual effort than earlier attack cycles required. That attack surface is real, expanding, and does not require defeating SHA-256 or secp256k1.

Why the Framing Debate Shapes Security Priorities

How the industry frames the AI-crypto security question shapes where resources go. If AI hacking fears channel attention toward quantum-resistant cryptography upgrades for blockchain base layers, that is a long-horizon engineering project with genuine merit. Standards bodies are already moving on that front: the US National Institute of Standards and Technology finalized its first post-quantum cryptography standards in 2024, though migrating systems as widely deployed as TLS and blockchain signatures is a multi-year undertaking. But if the same fears crowd out investment in wallet UX security, phishing-resistant authentication for exchanges, and user education around AI-generated scams, the tradeoff could leave users more exposed on the vectors that are actually active today.

CZ's argument, taken at its strongest, is a corrective to misallocated anxiety. The cryptographic layer of Bitcoin and Ethereum is not the weakest point in the system; the human and interface layers are. That assessment is consistent with how most high-profile crypto thefts have actually unfolded: private keys compromised through phishing and social manipulation, not broken elliptic-curve math. Recent episodes bear that out — the roughly $600 million theft from the Ronin bridge in 2022 began with a spear-phishing attack on an engineer, and the record-setting Bybit breach of February 2025, the largest in the sector's history, was traced to a compromised signing interface rather than any failure of cryptographic primitives.

The debate also carries implications for how AI agent infrastructure in crypto develops. As autonomous agents gain the ability to execute on-chain transactions, manage wallets, and interact with DeFi protocols, the attack surface expands at the application layer rather than the protocol layer. Securing those agent-to-chain interfaces against adversarial AI inputs is the more urgent engineering frontier, and it is distinct from the question of whether AI can crack Bitcoin's underlying cryptography. How exchanges, wallet providers, and agent developers split resources between base-layer cryptographic upgrades and interface hardening will be a practical indicator of which framing the industry actually adopts.

CZ's comparative framing does not close that conversation, but it recenters the debate on where the evidence of actual risk is concentrated: not in the cryptographic bedrock of public blockchains, but in the software, interface, and human layers that sit above them — the same layers where legacy financial systems have always been most exposed.

Disclaimer: This article is for informational purposes only and does not constitute financial or investment advice. Cryptocurrency and digital asset markets carry significant risk. Always do your own research before making decisions.