Vitalik Buterin rejects AI-driven Bitcoin crash prediction, argues crypto can withstand AI attacks

Ethereum co-founder Vitalik Buterin has pushed back against predictions that artificial intelligence could trigger a 50% Bitcoin crash within two years.

The debate began after Silicon Valley investor and AI-risk commentator Liron Shapiro said he sees a 50% probability that Bitcoin will fall more than 50% over the next two years because AI undermines what investors believe are the network’s security and robustness guarantees.

However, Buterin took the other side of this position, saying:

Buterin argued that Bitcoin should be able to handle security failures that do not require social consensus, including attacks affecting clients, mining pools and other network infrastructure. He also placed a very low probability on AI producing a fundamental break in Bitcoin’s hashing functions or proof-of-work system.

He said he would normally offer Shapiro a wager on the outcome, but his existing crypto holdings already amount to effectively making that bet with roughly 90% of his net worth. Buterin also suggested the same argument should apply to Ethereum.

The disagreement captures a growing divide over what increasingly capable AI systems mean for crypto security.

Shapiro’s argument does not require AI to break Bitcoin’s underlying cryptography. A wave of attacks that exposes weaknesses investors previously assumed were difficult or impossible to exploit could be enough to damage confidence, even if developers eventually repair the vulnerabilities.

Buterin is making a different calculation. He expects most AI-driven security failures to remain fixable and believes defenders will eventually benefit from the same technological advances attackers are using.

AI attacks are already testing crypto defenses

Evidence from across the crypto industry suggests the threat behind Shapiro’s warning is already becoming operational, even without a breakthrough against Bitcoin’s underlying cryptography.

In August, Bitcoin swap provider Boltz indefinitely suspended its service after months of automated, AI-assisted probing led to several contained exploits and began moving faster than its small development team could patch vulnerabilities.

Boltz said its non-custodial architecture protected customer funds, but the company absorbed losses from the exploits and eventually concluded it could no longer safely run the service.

The episode showed how AI could pressure the infrastructure surrounding Bitcoin without compromising proof-of-work or breaking the network’s cryptography. Attackers instead used automation to repeatedly search for weaknesses faster than defenders could investigate, patch, and deploy fixes.

That is the type of threat security executives expect to spread as offensive AI capabilities improve.

Deddy David, chief executive of blockchain security firm Cyvers, previously told CryptoSlate that the financial exposure from AI-powered crypto attacks could eventually reach hundreds of millions or billions of dollars.

“If AI can identify vulnerabilities at scale across core internet infrastructure, crypto will be one of the first markets to feel the impact,” David said.

The risk extends across wallets, bridges, exchanges, smart contracts and the software used to operate blockchain networks. Those systems provide attackers with considerably more potential entry points than Bitcoin’s core consensus mechanism itself.

The distinction is central to the disagreement between Buterin and Shapiro, as AI does not have to break SHA-256 to cause serious damage across the Bitcoin economy.

The unresolved question is whether those attacks remain problems developers can contain through upgrades and stronger defenses, as Buterin expects, or become severe enough to undermine the security assumptions investors have priced into Bitcoin.

Crypto and tech firms are racing to arm the defenders

The response from the broader technology industry increasingly resembles an arms race to ensure defensive AI develops at least as quickly as offensive capability.

Anthropic has restricted public access to its Claude Mythos model because of its ability to autonomously discover and weaponize software vulnerabilities.

The company has instead directed those capabilities toward defenders through Project Glasswing, an initiative involving technology companies and financial institutions including Amazon Web Services, Google, Microsoft and JPMorgan Chase.

The program uses Mythos Preview to identify and repair vulnerabilities in critical systems before malicious attackers equipped with similarly capable models can find them.

Anthropic has committed as much as $100 million in usage credits to the effort.

Meanwhile, the defensive push has since widened beyond individual companies.

More than 100 organizations, including Google, Microsoft, Anthropic and OpenAI, have signed an open letter warning governments and companies that AI-enabled cyberattacks are likely to become substantially more widespread and sophisticated within months.

Banks, payment companies, and major technology firms, including Capital One, Mastercard, Visa, Adobe, Oracle, and IBM, also joined the call.

The signatories argued that existing cybersecurity practices will not be sufficient as AI systems become more capable, particularly across historically under-resourced critical infrastructure.

They called for governments and technology companies to provide advanced defensive AI and security testing to organizations including hospitals and water utilities.

Crypto faces the same race, but with an additional complication: many of its systems hold immediately transferable financial assets and operate continuously on publicly visible infrastructure.

That makes the dispute between Buterin and Shapiro difficult to settle simply by asking whether AI can “break Bitcoin.”

The more immediate question is whether attackers gain a decisive advantage during the period before defensive tools, auditing practices and network infrastructure catch up.