Bitcoin unlikely to crash 50% on AI risks: Buterin says

Ethereum co-founder Vitalik Buterin rejected a Sept. 7 prediction that artificial intelligence could undermine confidence in Bitcoin’s security and cause its price to fall by more than 50% within two years.

Investor and AI-risk commentator Liron Shapira assigned a “50% confidence” level to his prediction that Bitcoin would suffer such a decline because AI could weaken what users consider the network’s security or robustness, according to his original post.

Buterin said he took “the opposite side” of that forecast. His response focused on Bitcoin’s technical adaptability rather than its market price, arguing that most AI-related threats could be addressed without changing the network’s core social agreement.

I take the opposite side of that.

My basic reasons are that I am quite optimistic about cybersecurity in the long term and I see the primary problem as being getting the transition, and I expect BTC to handle at least any issues that do not require social consensus well…

— vitalik.eth (@VitalikButerin) September 7, 2026

Buterin sees transition risk rather than cryptographic collapse

Buterin said he remains optimistic about cybersecurity over the long term. In his view, the main risk lies in managing the transition to a world where attackers and defenders both use increasingly capable AI systems.

He separated operational attacks from failures of Bitcoin’s underlying cryptography. Network-layer exploits, client vulnerabilities and attacks against mining infrastructure could require software updates, but they would not necessarily invalidate Bitcoin’s proof-of-work design.

Bitcoin’s network consists of several layers. Full-node software verifies transactions and blocks. Mining pools coordinate hash power, while miners provide the computing equipment that performs proof-of-work calculations. Internet infrastructure connects those participants.

AI could help attackers identify flaws in node software, wallet code, mining-pool systems or communications infrastructure. Defenders can also use AI to review code, test updates and detect suspicious behavior.

Buterin argued that client developers and mining pools could respond to those attacks through ordinary upgrades. Such changes would be less difficult than an emergency alteration of Bitcoin’s monetary rules or transaction history.

He described the probability of “actual breaks on hashes or PoW” as “tiny.” The statement is an assessment rather than a measurable guarantee, and it does not cover every cybersecurity risk affecting Bitcoin users.

Bitcoin’s proof of work is not the likeliest AI target

Bitcoin mining depends on SHA-256, the hash function used to process block headers and prove that miners performed computational work. A direct cryptographic break would require an attacker to find a practical weakness that sharply reduces the work needed to produce valid hashes.

However, Current generative AI systems have not demonstrated that capability. AI can improve software analysis and automate attack discovery, but it does not automatically defeat established cryptographic functions. An attacker also could not take control of Bitcoin simply by producing persuasive text, malicious code suggestions or automated social engineering. Compromising wallets, exchanges or individual developers would create losses, but it would not necessarily compromise Bitcoin’s consensus rules.

The more realistic near-term risks involve surrounding infrastructure. AI agents may improve phishing, credential theft, malware development and vulnerability discovery. Mining pools, exchanges, wallet providers and node operators could face more automated attacks.

A July security incident showed the distinction. Bitcoin Optech reported that a weakness in some COLDCARD-generated wallets produced insufficient entropy and exposed affected funds to theft. The problem involved wallet key generation rather than Bitcoin’s hash function or proof-of-work mechanism.

The technical report estimated losses exceeding 1,000 BTC at the time. Users of affected devices were advised to move funds to wallets created with secure external entropy.

Such incidents can damage confidence and cause major losses while leaving Bitcoin’s underlying ledger operational. They support Buterin’s argument that ecosystem software and custody remain more immediate attack surfaces than SHA-256.

AI could accelerate both attacks and defensive reviews

Shapira’s forecast appears to rest on an imbalance between attackers and defenders. Advanced AI could discover vulnerabilities faster than open-source developers can examine, patch and distribute software.

Bitcoin’s public codebase gives researchers broad access, but it also allows attackers to inspect the same software. A sufficiently capable automated system could review large repositories, identify unusual interactions and test attack paths continuously.

The Bitcoin Policy Institute and more than 40 digital-asset organizations raised that concern in August. They asked leading AI laboratories to provide vetted open-source security teams with controlled access to advanced models.

The coalition argued that defenders need early access to the same tools that sophisticated attackers may use. Its proposed program included computing resources, secure testing environments and direct communication with AI laboratory security teams.

Block, Coinbase, Strategy, MARA, Galaxy, BitGo, Brink, Trezor and several Bitcoin development groups supported the appeal. Crypto.news reported that the coalition sought frontier AI access for Bitcoin security researchers before more capable models became widely available.

A separate assessment from Bitcoin Policy Institute researcher Zack Shapiro and security specialist Efrat Fenigson’s coverage warned that the next several years could present a difficult transition. The concern was not that AI would directly solve Bitcoin’s proof of work, but that attackers could gain access to capable security tools before defenders.

As crypto.news previously reported, researchers warned that Bitcoin developers could fall behind AI-assisted attackers without comparable models, funding and computing access.

Social consensus would matter after a deeper failure

Buterin distinguished upgradeable operational problems from attacks that require social consensus. Bitcoin nodes and mining pools can adopt patched software when developers identify a conventional vulnerability.

A deeper cryptographic failure would be more difficult. If an attacker could forge digital signatures or bypass proof-of-work requirements, developers might need to introduce new cryptographic standards and coordinate a broad network migration.

Node operators, miners, exchanges, custodians and users would then need to agree on which software and transaction history to recognize. Disagreement could split the network or delay protective action.

Quantum computing is frequently discussed in this context because a sufficiently capable machine could threaten the elliptic-curve signatures protecting some Bitcoin holdings. That issue differs from Shapira’s AI claim because quantum computers use different computing methods rather than machine-learning techniques.

Crypto.news has reported that developers are exploring quantum-resistant infrastructure and migration tools before practical attacks emerge. Estimates about when such systems could threaten Bitcoin remain uncertain.

Buterin did not claim Bitcoin could respond easily to every cryptographic break. His argument was that those fundamental failures have a very low probability, while more plausible network attacks can be addressed through coordinated software changes.

Bitcoin showed no reaction tied directly to the debate

Bitcoin traded near $79,590 on Sept. 7, down approximately 0.5% during the latest session. Its intraday range was roughly $79,460 to $80,494.

There was no evidence connecting that movement to the exchange between Shapira and Buterin. Bitcoin’s price responds to liquidity, derivatives positioning, exchange-traded fund flows, macroeconomic data and broader risk sentiment.

Shapira’s prediction also does not specify a technical attack, affected software component or sequence connecting AI adoption to a 50% market decline. It is a probabilistic forecast and cannot be treated as a verified security assessment.

Buterin similarly offered no formal risk model supporting his confidence. His response provides a technical counterargument: Bitcoin can patch many operational vulnerabilities, while a direct break of its hash function or proof-of-work mechanism remains unlikely.

The dispute therefore concerns the speed and difficulty of Bitcoin’s security transition rather than evidence that AI has already compromised the network.