Has Claude become a useful tool for bitcoin developers?

[Antidote47k]

I have noticed Claude coming up more often in discussions surrounding bitcoin development and technical works.
My question is, for those who are actively building, reviewing code or contributing to Bitcoin-related projects, how are you actually finding it in practice?

Does it actively help in undertaking tasks like understanding bitcoin core, reviewing BIPs, debugging or general security research? Or is it still more useful in the aspect of explanation and exploration rather than serious development work?

Generally I’m really interested in hearing how people are integrating it into their workflow.

[NotATether]

Never use AI to operate wallets or to handle pull requests / issues on sensitive codebases such as Core.

Linux kernel is a bit different since the worst that can happen is crashes, but on Bitcoin Core, the code is money. So this is why such projects should stick with mainly human reviewers while using AI to help understand the prognosis and assist in creating bugfixes only.

[BlackHatCoiner]

I used it to draft this testnet softfork proposal and to write the code for it. I also periodically use it to write custom scripts that allow me to calculate useful info, such as for observing the sustainability of the network overtime.

Yes, it's capable of undertaking pretty serious tasks, there just needs to be someone responsible to prompt it right, and review the output.

[gmaxwell]

I think that using closed hosted LLMs is not very compatible with the ethos of Bitcoin.  In addition the the direct effect of any work done historically any work w/ bitcoin has prepared you to do *future* work as necessary.  But if you work using a closed hosted LLM you only get the direct benefit and potentially get worse at actually doing the work-- you become dependent.

You also allow unknown third parties realtime remote access to your systems and the ability to effectively impersonate you.

If you're using it on a system with your wallet(s) vs development in a sandbox a hosted LLM absolutely blows your privacy because the providers will log and retain all information that goes in and out.  I know from first hand experience that commercial LLM providers retain data far beyond what you would expect from their public statements and will provide it in response to absolutely anyone carrying a subpoena.

I would suggest as an alternative using Qwen-3.6-27B-- many regard it as competitive with Opus 4 and it has open weights and can be run on a workstation class gpu.  Even if you currently don't have the resources to run it locally using a hosted copy at least improves the autonomy since you can run locally if push comes to shove and you have many choices of where to host.

I can say first hand that Qwen running with a vm-sandboxed hermes is capable of very sophisticated development work.

[vapourminer]

I would suggest as an alternative using Qwen-3.6-27B-- many regard it as competitive with Opus 4 and it has open weights and can be run on a workstation class gpu. 

a >= 24 gb consumer class gpu (AMD 7900XTX, Nvidia 4090/5090) will run  qwen 3.6-27b q4 locally too. but the context window may not be enough for really big tasks if youre used to workstation stuff.

[dkbit98]

Claude can only be useful to check the code and find some hidden bugs and flaws in the code.
Someone already used this tool to exploit some shitcoins, and I am sure hackers are using it steal coins from exchanges.
But totally replacing any human bitcoin developer with Claude or any other AI would be stupid and dangerous.

[suzanne5223]

Claude can only be useful to check the code and find some hidden bugs and flaws in the code.
Someone already used this tool to exploit some shitcoins, and I am sure hackers are using it steal coins from exchanges.
But totally replacing any human bitcoin developer with Claude or any other AI would be stupid and dangerous.

I don't think Claude can be utilized for the things you are saying since, as soon as they are added, you will receive a response from Claude because it might negatively impact the AI's reputation, and I'm confident that Claude won't be involved in anything that would lead to criminal activity.  And from what I understand, developers can benefit from it, but not in the way you truly believe.

What s/he said is true.
I read on Malwarebytes news blog months ago about how some cybercriminal was able to use cutting-edge prompting with jailbreaking to bypass Claude’s protection barriers to use the AI to design vibe coding and launch attacks.
Although this might affect the reputation of AI, as you said, but it is what some people will always abuse innovative technology.

[Italian Panic]

In my experience, I wouldn’t rely on an AI or Claude (although I do recognise that Anthropic’s is the best on the market) because I have found it tends to be rather haphazard when looking for solutions. However, I have found it to be an excellent tool for code review, though you still have to write the code yourself. At most, you can get it to merge various pieces of code, but never start from scratch! The initial structure must always be your own.

[Mahiyammahi]

I think that using closed hosted LLMs is not very compatible with the ethos of Bitcoin.  In addition the the direct effect of any work done historically any work w/ bitcoin has prepared you to do *future* work as necessary.  But if you work using a closed hosted LLM you only get the direct benefit and potentially get worse at actually doing the work-- you become dependent.

You also allow unknown third parties realtime remote access to your systems and the ability to effectively impersonate you.

If you're using it on a system with your wallet(s) vs development in a sandbox a hosted LLM absolutely blows your privacy because the providers will log and retain all information that goes in and out.  I know from first hand experience that commercial LLM providers retain data far beyond what you would expect from their public statements and will provide it in response to absolutely anyone carrying a subpoena.

So even trying to build a simple contracts would be at risks for getting exploited? As for storing data I hope they don't sell our data. If we take accountable these thing almost everything would be impossible to build with AI and scale.

I would suggest as an alternative using Qwen-3.6-27B-- many regard it as competitive with Opus 4 and it has open weights and can be run on a workstation class gpu.  Even if you currently don't have the resources to run it locally using a hosted copy at least improves the autonomy since you can run locally if push comes to shove and you have many choices of where to host.

I can say first hand that Qwen running with a vm-sandboxed hermes is capable of very sophisticated development work.

Local LLMs can save our token consumption and ensure us safety. But to run a compatible LLM requires more GPU which is usually costly. If I wants to run a compatible AI model which would require me 5$/month it would requires me atleast $3k built. So as a consumer level it's expensive.

Do you think we can still use those commercial AI models in any alternative way without having the risks of our data leak and we can make sure our atmost security on our projects.

[pooya87]

AI is definitely a useful tool for coders (a lot of other tasks) but it is important for us to first understand what type of a tool it is. The problem is in the name: A.I. which they chose for marketing purposes. But AI is not intelligence. It cannot think. It cannot reason and most importantly it cannot innovate. A.I. is best described as a statistical pattern matcher. But that doesn't sell! The companies couldn't say "come invest millions in our 'statistical pattern matcher'"

After we get that definition correctly, we can move on to utilize this tool more efficiently. Then you wouldn't expect a "pattern matcher" to write you an innovative code solving a new problem. You'd use it to solve a problem that was solved a million times before my many others and that AI has it in its database aka training data.

My question is, for those who are actively building, reviewing code or contributing to Bitcoin-related projects, how are you actually finding it in practice?

I have only used them for code review and have already found one serious bug (a dumb IndexOutOfRange bug) and a bunch of small improvements in my code.

[dzonikg28]

AI is definitely a useful tool for coders (a lot of other tasks) but it is important for us to first understand what type of a tool it is. The problem is in the name: A.I. which they chose for marketing purposes. But AI is not intelligence. It cannot think. It cannot reason and most importantly it cannot innovate. A.I. is best described as a statistical pattern matcher. But that doesn't sell! The companies couldn't say "come invest millions in our 'statistical pattern matcher'"

After we get that definition correctly, we can move on to utilize this tool more efficiently. Then you wouldn't expect a "pattern matcher" to write you an innovative code solving a new problem. You'd use it to solve a problem that was solved a million times before my many others and that AI has it in its database aka training data.

My question is, for those who are actively building, reviewing code or contributing to Bitcoin-related projects, how are you actually finding it in practice?

I have only used them for code review and have already found one serious bug (a dumb IndexOutOfRange bug) and a bunch of small improvements in my code.

This is a good explanation of how to understand A.I., especially when you say it is not intelligent in the sense that humans are intelligent, like connecting the dots and add synergetic value beyond just creating patterns. But this brings me the following question: the term "statistical pattern matcher" makes a lot of sense to me, but that does not exclude the possibility that A.I. extrapolates existing patterns into potential new patterns or solutions, does it? What I frequently realise is that you can challenge A.I. answers to many questions, especially opinions because it seems to scrape the Internet and present something from there as its own, but then corrects it when you challenge it.

As for solutions in programming, what is your opinion on OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack? I don't have a strong opinion on that because I am not very familiar with programming, but maybe you can give yours. I was thinking whether this could as well be a marketing stunt? Would a good hacker have found this way to get into the system as efficiently? Like within the same time window? Was it a sophisticated attack that the A.I. conducted? Or you think this incident was hyped up far beyond what it really was?

If it was, isn't that an indication that A.I. will soon be able to write good human level software?

[ABCbits]

Do you think we can still use those commercial AI models in any alternative way without having the risks of our data leak and we can make sure our atmost security on our projects.

Whether we talk about AI or other digital service, privacy and data security usually comes down to
1. Their privacy policy, such as what kind of data stored, processed or shared.
2. Privacy and data law either on their or your country.
3. You trust them or not.

As for solutions in programming, what is your opinion on OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack? I don't have a strong opinion on that because I am not very familiar with programming, but maybe you can give yours. I was thinking whether this could as well be a marketing stunt? Would a good hacker have found this way to get into the system as efficiently? Like within the same time window? Was it a sophisticated attack that the A.I. conducted? Or you think this incident was hyped up far beyond what it really was?

I don't know the technical detail. But i believe it's just another combination of public stunt and lie in to convince government to ban their competitor. Despite name of their company, they are against open AI.

The impressive capabilities of Chinese lab Moonshot’s Kimi K3, the biggest open-weight large language model, have kicked off a debate that conflates two things: the economic possibilities of American AI giants and the future of LLMs as a technology.

OpenAI’s head of strategic futures, Dean W. Ball, went so far as to argue that the U.S. government should find a pretext to create regulatory fear, uncertainty, and distrust around the new models, since open-weight models must necessarily deter capital spending by the frontier labs.

His X post also use some obvious attack, such as calling open AI as communism.

4. One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes:

I believe OpenAI behavior is similar with how Microsoft openly attack linux and other open source decades ago.

MS ANALYSTS Steve Ballmer was the only person to raise the issue of Linux when he wrapped up Microsoft's annual financial analysts meeting in Seattle, although he put Sun and Oracle ahead in terms of being stronger competitors. They of course are 'civilised' competitors - but the Linux crowd, in the world of Prez Steve, are communists.

[bitmover]

AI is definitely a useful tool for coders (a lot of other tasks) but it is important for us to first understand what type of a tool it is. The problem is in the name: A.I. which they chose for marketing purposes. But AI is not intelligence. It cannot think. It cannot reason and most importantly it cannot innovate. A.I. is best described as a statistical pattern matcher. But that doesn't sell! The companies couldn't say "come invest millions in our 'statistical pattern matcher'"

After we get that definition correctly, we can move on to utilize this tool more efficiently. Then you wouldn't expect a "pattern matcher" to write you an innovative code solving a new problem. You'd use it to solve a problem that was solved a million times before my many others and that AI has it in its database aka training data.

My question is, for those who are actively building, reviewing code or contributing to Bitcoin-related projects, how are you actually finding it in practice?

I have only used them for code review and have already found one serious bug (a dumb IndexOutOfRange bug) and a bunch of small improvements in my code.

I agree.

I think that just prompting and asking the IA to deliver a full code/app for you won't work in the long run. AI will make mistakes and the app will have a lot of bugs.

But it is useful to make me code faster. I usually ask AI tasks like this: "create a function A that will organize this data into an array and then loop every element using this other function B, and finally organize the response into C". Then I can carefully look at what it did and make just minor adjustments. And this actually saves a lot of time.

I am not a core developer, I just AI to solve problems that have been solved already a million times... But I think a core developer might have some use for it, specially if running locally to protect his privacy and security

[Synchronice]

What s/he said is true.
I read on Malwarebytes news blog months ago about how some cybercriminal was able to use cutting-edge prompting with jailbreaking to bypass Claude’s protection barriers to use the AI to design vibe coding and launch attacks.
Although this might affect the reputation of AI, as you said, but it is what some people will always abuse innovative technology.

Recently, Claude announced Opus 5.0 and already made available for everyone. As they claim, Opus 5.0 is as powerful as Fable 5 at 1/2 cost of Fable. If someone is able to bypass Claude's restriction to launch attacks, then it's really very interesting how powerful Opus 5.0 and Fable will be in this case, especially if we keep in mind how big the hype was around Mythos 5 and as they claim, Fable 5 is mythos with restrictions.

Another interesting thing for me is, what if someone buys a powerful combination of GPU, RAM and other components and downloads currently the best open-source AI model. Can such model be used for launching cyber attacks? I don't know and would really love to get the answer. I'm also really very curious about what will be the capability of Opus 6 and Fable 6, next gen Claude AI models.

Also, the way ChatGPT improved their AI with the release of 5.5 makes me think that they might drop a better model than Claude Opus/Fable 5 is.