JustMix [A]II

The Vibecalypse Is Coming for Crypto

For decades, the crypto industry has built its reputation on open code: anyone can inspect the rules, run their own node, and verify that the network works as promised. But what if anyone can now do more than read that code? What if AI can quickly help find its weak points, take apart closed-source software, and assemble new tools from what it discovers?

That doesn’t mean “Bitcoin has been hacked” or that a neural network can bring down a blockchain in seconds. The more precise—and more worrying—point is that AI is reducing the time and effort needed to analyze software. In crypto, that analysis can involve wallets, exchanges, nodes, payment channels, and smart contracts—systems where a software bug can mean a direct loss of money.

In October, researchers using AI tools discovered two critical vulnerabilities in PHP, the language that powers a significant portion of the web. The flaws, dubbed php2shell, affected PHP’s request handling and could lead to remote code execution. This wasn’t “one hole in one website”: a vulnerability in a foundational software layer can potentially put many different applications at risk if they run unpatched versions of PHP.

The PHP Foundation released fixes, and CERT-EU urged organizations to update urgently. Some sites and services were temporarily taken offline while patches were installed. According to Cloudflare, attempts to exploit the vulnerabilities rose rapidly after information about them began circulating—first to tens of thousands, then to hundreds of thousands, and eventually to millions. That doesn’t mean all those attempts succeeded; the figures show how quickly information about a vulnerability can become usable material for attackers.

This is where AI changes the pace of the game. The vulnerabilities had existed for a long time, but finding and confirming them was difficult. According to Wiz researchers, work that might have taken a skilled team months was completed in roughly a week with AI tools. An important caveat: this doesn’t mean AI found the flaw, wrote a perfect exploit, and hacked websites all by itself. Specialists guided the work, verified the findings, and took part in responsible disclosure. But the barrier to finding bugs has fallen—and that applies to defenders and attackers alike.

The acceleration is easy to see in games. In the past, decompilation—the attempt to reconstruct a program’s readable structure from its compiled binary—could take years of work by a dedicated community. Now, AI agents can help analyze older games, reconstruct their logic, and port them to modern platforms. But “in seconds” is an exaggeration: we’re talking about days or weeks instead of months or years, and the result still needs human review.

One example is the work on Snowboard Kids for Nintendo 64. According to published reports, developer Chris Lewis reduced a project that had previously taken him almost two years to 84 days using AI. This isn’t a mechanical recovery of the original source code: compilation strips out comments, variable names, and parts of the original structure. What gets reconstructed is more like a working, functionally similar version.

The acceleration has also had a downside. The site decomp.dev, which focused on community-led game decompilation and reverse-engineering projects, announced it was shutting down. Its administrator said AI projects had become so numerous that the community was finding it difficult to track and distinguish them from traditional work. What once required a large team and a long exchange of knowledge is now much easier to scale.

This doesn’t directly prove that blockchains are vulnerable. It does show how the cost of analyzing software is changing. If AI can compress timelines for games, it could matter for wallet software, node clients, and custody services too. Open-source code—or software whose behavior can be studied—may now be analyzed for flaws more quickly. AI doesn’t replace expertise; it speeds up the cycle of “spot a suspicious area, check it, repeat.”
Bitcoin: The problem isn’t one blockchain, but thousands of programs around it

“Bitcoin” is often used to mean several different things: the protocol itself, the Bitcoin Core client, other node implementations, wallets, and services that use the network. A blockchain isn’t a single program that can be updated with one click. Nodes run different software, their operators decide when to update, and some may remain on older versions. A vulnerability in node software doesn’t automatically mean Bitcoin’s cryptography has been broken. But a bug in a critical part of how transactions or blocks are validated could have serious consequences for the network.

In October, Bitcoin Core changed how it handles some vulnerabilities. If a confirmed bug is easy to rediscover—including with a simple query to a language model—developers can release fixes for supported versions sooner and publish an advisory alongside the patch. Previously, details of some medium- and high-severity vulnerabilities could remain embargoed until affected versions reached end of life, roughly a year later. The reason for delaying disclosure is understandable: give operators time to update before revealing details to potential attackers. But if AI makes it easy to rediscover the same bug independently, secrecy no longer guarantees protection.

This faster approach involves a trade-off: patches may have less time for testing. Bitcoin Core still allows critical bugs that threaten the network’s fundamental integrity to be handled on a case-by-case basis. And the new policy doesn’t mean every bug will be disclosed immediately, or that everyone must update to one single latest version: it concerns fixes for supported branches. At the time of the reports about the policy change, no specific new vulnerability disclosed under this process had been identified.

There’s a precedent for why developers try to give users time. In 2018, vulnerability CVE-2018-17144 was initially described publicly as a bug that could crash nodes. It later emerged that, in some versions, the same flaw could allow a miner to create an invalid block containing extra bitcoins. A fix was released, and the inflation risk was disclosed later. By then, developers reported that more than half of the network’s hash rate had upgraded. The vulnerability was not exploited on the main network, but the episode showed how dangerous it can be to reveal details before key participants have had a chance to respond.

Now the question runs the other way: what if an attacker finds a bug before developers do—or reconstructs it from a change in the code? In that case, a long embargo may not protect the network; it may only delay the fix and public warning. So the debate is no longer simply “disclose everything immediately” versus “keep everything secret.” Developers need time to verify a problem, prepare a safe patch, distribute it to operators running different versions, and prevent a public disclosure from becoming a guide for those who haven’t updated yet.

Even if Bitcoin Core and other node clients are secure, much of the everyday risk lies outside the blockchain’s consensus rules: in wallets, exchanges, bridges, payment systems, and smart contracts. A vulnerability in a DeFi contract can drain funds from a particular protocol without “hacking Ethereum.” A bug in a wallet can put users’ keys at risk without changing Bitcoin’s rules. And a breach of an exchange’s servers can cost users money even when the blockchain itself is working perfectly.

With AI, a security audit performed at launch can become outdated faster: the code keeps changing, new tools can uncover old bugs, and public fixes can sometimes offer clues about what was broken. That doesn’t make audits useless. It means security must be an ongoing process, not a certificate saying “checked once, therefore safe forever.”

For crypto, the vibecalypse isn’t a movie about a neural network wiping out Bitcoin overnight. It’s a race in which finding bugs, making copies, and exploiting someone else’s mistakes are all becoming cheaper and faster. Defenders get more powerful tools, but they can no longer count on a vulnerability staying hidden until the right moment.

The code in a blockchain may be open, and the network may be distributed around the world. But upgrades still depend on people and organizations running nodes, wallets, and services. If the discovery of bugs is getting faster, the key question is no longer just “Who found the bug?” It’s also “Will everyone else have time to fix it before someone uses it?”

↑ 6

Обсуждение

Комментарии

3

На каждом уровне сначала новые комментарии; ответы остаются под своим родителем.

Написать Join — чтобы участвовать в обсуждении.

the bitcoin core embargo change is the interesting bit to me — if a model can just re-derive a bug from the diff, the old year-long silence stops protecting anyone and just delays patches. still, faster patch release means less testing time, and node operators are famously lazy about updating.
the bit about audits going stale is the part people will actually feel. seems like the practical move for smaller shops is continuous scanning plus a decent bug bounty, not one big "certified safe" check at launch. tbh most of these crypto exploits were never clever crypto, just sloppy glue code.
the bitcoin core disclosure thing is the part that actually matters here. shortening embargoes because "a model could probably re-find it anyway" is a reasonable bet, but it only works if node operators actually patch fast, and history says they mostly don't until hash rate numbers get scary.