Cofundador do Ethereum (ETH), Vitalik Buterin, contesta o “fim” da cibersegurança com IA e mantém 90% do patrimônio em cripto

COINOTAG · 2026-09-17T10:30:47.000Z

• Vitalik Buterin rejeitou as alegações de que o hacking com IA vai condenar a cibersegurança em uma postagem de 17 de setembro no X. • Buterin disse que cerca de 90% do valor líquido dele permanece em cripto, uma aposta implícita em sistemas seguros. • Buterin executou localmente o Qwen3.8-Flash-Next da Alibaba, com 125 bilhões de parâmetros, em um laptop AMD Strix Halo. Verificação Formal, Não Desespero O cofundador do Ethereum (ETH), Vitalik Buterin, se opôs à visão cada vez mais comum de que a inteligência artificial capaz tornará a cibersegurança impossível de vencer. Ele argumentou, em uma longa postagem de 17 de setembro no X, que o resultado oposto é mais provável. “É uma interpretação cada vez mais comum de que hacking com IA significa que a cibersegurança está condenada. Eu discordo”, escreveu Buterin, afirmando que a IA pode, em vez disso, entregar aos defensores ferramentas para construir softwares que sejam significativamente mais difíceis de explorar desde o início. O núcleo do argumento dele é a verificação formal — demonstrar matematicamente que um programa, incluindo um contrato inteligente, satisfaz propriedades específicas. “Se a IA consegue provar Navier-Stokes e FLT, então a IA pode provar a afirmação ‘este programa é seguro’ como um teorema matemático”, escreveu ele, citando dois problemas famosos por serem especialmente difíceis em matemática. Ele reconheceu o problema de definição: mesmo determinar o que “seguro” significa para algo como o aplicativo de mensagens Signal pode ultrapassar mil linhas de especificação, considerando superfícies de ataque que vão desde mensagens forjadas e reencenadas até compiladores maliciosos e vazamentos no nível de hardware. Buterin também vinculou a tese à própria exposição pessoal, revelando que cerca de 90% do valor líquido dele permanece em cripto — uma aposta implícita, segundo ele, de que sistemas digitais suficientemente seguros podem ser construídos mesmo contra atacantes sofisticados.

• Vitalik Buterin rejected claims that AI hacking will doom cybersecurity in a September 17 X post.
• Buterin said roughly 90% of his net worth remains in crypto, an implicit bet on secure systems.
• Buterin ran Alibaba's 125-billion-parameter Qwen3.8-Flash-Next locally on an AMD Strix Halo laptop.
Formal Verification, Not Doom
Ethereum (ETH) co-founder Vitalik Buterin has pushed back against the increasingly common view that capable artificial intelligence will make cybersecurity unwinnable, arguing in a lengthy September 17 post on X that the opposite outcome is more likely. “It’s an increasingly common take that AI hacking means cybersecurity is doomed. I disagree,” Buterin wrote, stating that AI may instead hand defenders the tools to build software that is significantly harder to exploit in the first place. The core of his argument is formal verification — mathematically demonstrating that a program, including a smart contract, satisfies specific properties. “If AI can prove Navier-Stokes and FLT, then AI can prove the statement ‘this program is secure’ as a mathematical theorem,” he wrote, referencing two famously difficult problems in mathematics. He acknowledged the definition problem: even pinning down what “secure” means for something like the Signal messaging app can stretch past a thousand lines of specification, given attack surfaces ranging from forged and replayed messages to malicious compilers and hardware-level leaks. Buterin also tied the thesis to personal exposure, revealing that roughly 90% of his net worth remains in crypto — an implicit bet, he argued, that sufficiently secure digital systems can be built even against sophisticated attackers.
125 Billion Parameters on a Laptop
In a separate X post the same day, the Ethereum co-founder turned from security to local AI, posting speed results from running Alibaba’s Qwen3.8-Flash-Next entirely on his own laptop. Unlike ChatGPT, the setup never contacts a cloud server — the model sits on the machine, and the machine answers by itself. The hardware matters: his laptop uses AMD’s Strix Halo chip, which places the processor and graphics half on a single piece of silicon sharing one memory pool of up to 128 gigabytes, against the 8 to 24 gigabytes typical of a discrete graphics card. That unified design is what lets one laptop hold a model that until recently needed server hardware. Alibaba published the open weights on August 26; the model holds 125 billion parameters yet activates only 6 billion at a time, keeping memory demands modest. Buterin called the model “truly impressive” and noted that llama.cpp has been rapidly getting better at processing it. The posted speeds were practical for ordinary work — short prompts returned at a comfortable reading pace, though output slowed once prompts ran to tens of thousands of words, leaving very long documents as the weak spot.
Privacy Without the Cloud
Beyond raw speed, Buterin framed local inference as a privacy shield. A local model answers on the device, so no provider ever receives the request. For demanding work, he proposed a split: “use your local model to orchestrate queries to powerful models so your queries don’t leak your personal information.” In practice, the local model would strip names, wallet addresses or private code out of a prompt and pass along only the remaining question — cutting what leaves the device, though not guaranteeing that nothing sensitive slips through. The pitch matches his record: he warned about surveillance during the EU chat control fight, and crypto users have pushed for tighter limits on automated agents for similar reasons. A class action filed in May accuses OpenAI of sharing ChatGPT user queries with Meta and Google. On-chain security incidents, such as the recent $7.8M Ethereum (ETH) Safe wallet exploit, keep reinforcing that caution, while self-sovereign tooling like .ETH domain names already lets users hold identity outside centralized platforms. Cloud providers still own the frontier, but every gain in local performance moves routine work off their servers.
Security as the Common Thread
Our reading of the two primary posts, both published on Buterin’s own X account on September 17, is that they form one argument: AI is dual-use, and the side that formalizes its reasoning first wins. Formal verification could make validator software and contract code provably sound across the broader Ethereum ecosystem, while local models keep sensitive data off third-party servers. With roughly 90% of his net worth held in crypto, Buterin’s confidence in proof-of-stake security is more than academic — it is a priced-in conviction.