Latest

Artificial Intelligence

Vitalik Buterin proposes AI as the definitive auditor of open source software

Vitalik Buterin discusses how AI can help develop Ethereum and open source software in general, but not everything is so perfect.

Vitalik Buterin's prophecy: AI as the ultimate auditor of open source software
Vitalik Buterin's prophecy: AI as the ultimate auditor of open source software

The relationship between open source software and artificial intelligence is today crossed by a fundamental tension, one that defines the future of open technology. On the one hand, there is a legitimate fear within the community that AI will consolidate itself as a centralized, opaque black box controlled by a few corporate actors. A situation that undermines the transparency and reliability of the code of such projects.

And, on the other hand, open source projects (including cryptocurrency projects) find themselves at a turning point, facing a critical need for advanced tools capable of accelerating technical development without compromising the integrity or security of those developments.

AI and open source software

Faced with this dilemma, Vitalik Buterin has recently suggested that the key to resolving this paradox does not lie in resisting technological change, but in knowing how to direct the immense power of AI toward strengthening trustlessness. This concept, which represents the absence of any need to trust third parties, stands as the fundamental pillar of crypto philosophy and, of course, of the open source world in general.

And the message is simple: code security is the only valid law. Thus, by transforming artificial intelligence from a potential threat into the ultimate auditor of open source software, the ecosystem can aspire to a standard of security and transparency that until recently seemed unattainable.

JACK'S SINCERICIDE: HE FIRES 4,000 HUMANS AT BLOCK BECAUSE AI IS CHEAPER

The Ethereum experiment

Faced with this reality, Ethereum is serving as the main laboratory for this integration, demonstrating that what once seemed unrealistic is now a clear technological direction toward the automation of development.

Recently, it became possible to experiment with the full roadmap for Ethereum through 2030, in a matter of weeks using vibe-coding, a milestone that evidences a massive acceleration in code writing.

This trend is so powerful that tools running locally on laptops now make it possible to replicate complex software, such as blogging systems, in a fraction of the time it took just six months ago.

A measured risk?

However, this vertiginous speed carries crucial warnings about the technical integrity of projects. Building this way, especially without having previously defined Ethereum Improvement Proposals (EIPs), inevitably generates critical errors and stub or incomplete versions, where artificial intelligence omits the real complexity of the system or simply does not attempt to produce the full version.

Basically, AI accelerates development, but if you haven't thought everything through properly, what it generates may be neither functional nor secure, or outright garbage. And that is where the great challenge lies: balancing the use of AI, human invention and lateral thinking, and thus managing to develop complete and functional ideas.

However, despite the risk of this initial experimentation, what truly matters is the trajectory of change; vibe-coding is only the prelude to a future where development and bug fixing will occur on a drastically reduced timescale, with humans being a vital part of reaching the goals. A measured risk, to accelerate software development.

Balancing the human-machine relationship

And for this to work, it is necessary to balance the scales of this love-hate relationship between humans and machines. Only then does a paradigm shift emerge in which artificial intelligence stops being merely a generator of lines of code and becomes the definitive auditor of open source software.

Buterin's central proposal consists of not using this technology exclusively to accelerate the writing of programs, but to audit and verify code massively and systematically. The ideal approach for the community is to split the benefits of AI equitably: take half the gains in speed to accelerate innovation and devote the other half to strengthening security through processes that were previously human and slow.

THE AI ECONOMY DEPENDS ON CRYPTOCURRENCIES TO FUNCTION AND COINBASE KNOWS IT

For example, in this new scheme, formal verification becomes the cornerstone for guaranteeing the integrity of decentralized networks. Initiatives such as @leanethereum are already demonstrating this potential by using artificial intelligence to generate machine-verifiable proofs of some of the most complex theorems on which the security of STARKs rests.

In this way, the development process becomes a quest for total security where AI acts as a tireless technical shield.

Toward error-free code?

But Buterin's vision does not stop there. Buterin states that this technological evolution marks the end of what many considered an idealistic illusion: the possibility of achieving truly error-free code.

Historically, perfection in software has been considered an unattainable fantasy, but today that vision seems to be changing radically. Although artificial intelligence does not completely eliminate the inevitable friction with errors and inconsistencies between implementations, it does make it possible to accelerate the review and audit process.

Despite this, it is essential to remain realistic about security, understanding that total protection remains theoretically impossible.

A new idealism

In fact, Buterin's vision generates a new idealism: AI can now help us generate secure code. And that, given the experiences so far, does not seem to be entirely true. The Moonwellhack, for example, was introduced by Claude, the most powerful AI for development. The cost of that error was 1.8 million dollars lost due to a bug generated by the AI.

Another example is seen at SaaStr, where Replit's AI did not follow the code freeze orders, thereby generating false data that affected the service.

The curl case and the misuse of AI in cybersecurity

But perhaps the most striking and dangerous example is seen in a small open source project: curl.

Daniel Stenberg, the lead maintainer of curl, which happens to be a critical tool for almost all modern software, has repeatedly denounced how AI is flooding the project with false or hallucinated security reports.

8M IN TARRAGONA LOOKS TO THE FUTURE: AI, BLOCKCHAIN AND FEMALE POWER

The error happens because contributors use AI tools to find vulnerabilities and submit reports that look legitimate at first glance but, after deep inspection, turn out to be technical slop. Thus, between 2024 and 2025 alone, the rate of valid reports in the project fell from acceptable levels to barely 5%.

The cost here is maintainer burnout. Evaluating each of these false reports takes between 10 and 15 minutes. If we consider that large projects receive hundreds of these per month, the cost in expert time (often valued at more than $150/hour) amounts to a loss of productivity of thousands of dollars per month that should be devoted to improving the real code. In fact, to avoid this, curl has closed its bounty program, and they are analyzing whether to reopen it with new guidelines or leave it closed permanently.

José Maldonado

Author

José Maldonado

Activist and blogger on technology, free software and blockchain.

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading