Bitcoin: Is AI More Dangerous Than Quantum? 2 Developers Explain Everything You Need to Know

AI Already Helps Find More Vulnerabilities

AI assistants are now used daily by developers to write, correct, or review code. This increase in productivity also affects cybersecurity, where the same models can multiply tests and search for vulnerabilities in open-source projects or digital security tools.

At the end of July, Anthropic's Frontier Red Team announced that Claude Mythos Preview had discovered new methods to attack several cryptographic algorithms.

Claude Mythos reportedly found a vulnerability in HAWK, a candidate for a future quantum-resistant signature standard. Despite two years of examination by human researchers, Mythos was able to find an attack in 60 hours that halved the effective robustness of its keys. The experiment reportedly cost around $100,000 in API usage.

Recent BTC thefts suffered by Coldcard users have nonetheless shown the consequences that an implementation error can have in the Bitcoin ecosystem.

A bug introduced in 2021 had reduced the entropy of certain seeds generated by several Coldcard models. According to several experts, the attacker may have used a LLM to generate the scripts that automated these thefts and could also have used it to discover the vulnerability.

To understand what this acceleration really changes for cryptocurrencies, Cryptoast gathered testimonies from Amira Bouguera, an Ethereum developer, and a Bitcoin Core developer who wished to remain anonymous.

On the Bitcoin Core side, this evolution is already visible:

Bitcoin Core developers are very aware of this evolution. There are more security reports, and many contributors are already using AI.

AI Mainly Amplifies the Developer's Capabilities

According to the Bitcoin Core developer interviewed, it would be a mistake to present AI as an autonomous intelligence necessarily superior to human researchers.

Someone who is not very experienced will simply produce 10 times more bad code, but a security expert can become 10 times more productive and find many more bugs.

He believes that AI primarily acts as a multiplier, allowing a competent developer to review more code, test more scenarios, and consider attacks previously deemed too unlikely or too costly.

The developer also estimates that current AI models have mainly brought "thickness rather than depth," increasing the volume of work that can be done but not necessarily improving its quality or relevance.

AI and Quantum Do Not Threaten Bitcoin in the Same Way

The risk posed by quantum computers to cryptocurrencies, as well as to the entire financial world, would thus be direct and potentially devastating. However, no computer capable of executing this attack exists today.

As for AI, the risk is already present. It does not allow for breaking ECDSA or Schnorr, but it can already accelerate the search for errors present in clients, wallets, cryptographic libraries, or future standards meant to protect Bitcoin from quantum threats.

Amira Bouguera, a cryptographer and protocol engineer interviewed by Cryptoast, thus dismisses any immediate threat related to Anthropic's announcement:

The risk to blockchains is virtually zero today, at least following this announcement [from Anthropic]. But it shows that things are moving very quickly. We need to start using these tools to find vulnerabilities before someone else does.

With HAWK, for example, AI did not use a quantum computer to attack Bitcoin. Instead, it was able to find a flaw in an algorithm specifically designed to protect digital systems against a future quantum threat.

The Human Bottleneck Remains

The increase in the capabilities of artificial intelligences does not only benefit attackers. On the developers' side, security researchers and Bitcoin defenders also have the same tools to examine the code and fix vulnerabilities.

<< It's not as if only attackers have more capabilities. Defenders have more too >>, emphasizes the Bitcoin Core developer.

The main imbalance could rather appear in the functioning of open-source projects, as an AI can generate a lot of code, open pull requests, or produce vulnerability reports in just a few minutes. Checking the relevance of proposed changes, measuring their consequences, and deciding whether they deserve the project's attention still takes time from experienced contributors.

The developer interviewed is not only concerned about an AI suddenly being able to break Bitcoin applications but also about the increasing noise around its development:

Where there is more asymmetry, it is on the social level. Maintainers risk being overwhelmed by requests for changes that are not priorities and divert attention from more important issues.

AI poses a more immediate threat than quantum computing, as its effects are already visible in vulnerability research and code production. However, at this stage, it is not more dangerous to the cryptography of blockchains.

A quantum computer capable of breaking Bitcoin signatures would represent a more direct threat, while AI accelerates a race in which attackers and defenders currently have the same weapons.