I have a similar claude story (much less money though), with the IRS R&D tax credit. The auditing firm initially said we qualify for $0. But then I had claude analyze past R&D reports and our expenses and it found the problem. The auditor had miscategorized our company.
So claude drafted an email even pointing to the right Internal Revenue Code (IRS Law), and specify why we fall under a specific category. The auditor got back to me two days later admitting their mistake and said our company now qualifies for $8k in tax credits. And a few months ago, it identified items in our AWS that saved us $250 a month (paying for itself).
So now I joke that even if I have a claude max plan, I've still come out ahead financially.
By my estimation (guess) you won't actually need to spend that much because the models are already getting a point where they don't need to get a whole lot better to be extremely helpful across many domains.
And it looks like those very helpful capabilities will continue to transfer to smaller models as well, as architectures and training regimes continue to refine.
I can fairly easily imagine a world where the only people needing to spend a lot of money on models are those that are using them to solve truly novel problems. The rest of us will get plenty of use at reasonable costs for the typical day-to-day helpful stuff.
To be fair regarding taxes it should be that you get all exemptions by default and the other side telling/justifying you why you don't qualify, instead of you not getting the things you should because you're not sure how to interpret the law or don't know this or that rule exists. Taxes shouldn't be that difficult, and the US version of it seems to be behind even the one I have here in europe (where my taxes are done "for me" for anything non business owner related). If the government is calculating your taxes anyway, they should just give the number to you instead of asking you a number and you better have the same as us or you're guilty of something.
Feels like a system that is deliberatly made to be more punishing for those who can't afford the help or the education to figure things out.
I would agree with you on most situations (like 1040 personal income taxes especially).
But in this particular tax credit, there's no way for the gov to know automatically what percentage of payroll was spent in qualified R&D expenses, since it's day to day business operations. Which is why we are _forced_ to hire an outside firm and pay them thousands of dollars (when Claude did an even better job), just to analyze how much of our time qualified as R&D expenses.
The problem I have is that I am forced to have to find a firm to do this, and most firms won't even work with companies as small as ours. So then we're stuck and losing out on years of R&D tax credits at the moment, when I really don't need them anymore, to be honest.
Claude Code is really good at stuff like this. The other day I tried to recover some images from an SD card that had gone bad. I used GetDataBack to recover files, but they appeared to be malformed and didn't open in image viewers.
I tasked Claude to analyze the files and figure out what's going on, and eventually we figured out that each file had a custom metadata header + thumbnail + actual image concatenated. I had it write a python script and was able to recover all the images with their metadata. It's nothing a human couldn't have figured out, but it was definitely WAY faster than doing it myself.
I've also used Claude in the past to figure out how to break into routers with locked down firmware. It's great at suggesting and trying approaches.
I have a friend that just picked up a new consulting job resurrecting an ancient Windows desktop application. No source control, no tests. And it's spread out over a dozen different folders with names like "_old", "_new" and "dates". Claude's doing a tremendous job in getting him to grips with what is actually happening in the application, what's relevant, what's not, what's different. I think it's literally saving him days and days at work.
if your friend has access to the binary and can pull it out to different box, they might get a lot out of a ghidra mcp -> https://github.com/LaurieWired/GhidraMCP
> I have a friend that just picked up a new consulting job resurrecting an ancient Windows desktop application. No source control, no tests. And it's spread out over a dozen different folders with names like "_old", "_new" and "dates".
That doesn't sound very impressive. Not being tracked with a version control system is fixed instantly with a git init, git add ., git commit .no AI required.
Covering the app with tests is also something that requires no AI. At most, coding agents can generate characterization tests in broad sweeps, but we are talking about a delta between hand rolling and vibe-coding of a couple of days.
Where LLM shines is helping developers build up an understanding of what is in place. Running /explain on a codebase can quickly provide you with a high level summary of what's in place.
I did EXACTLY that last night. Was doing by hand for about an hour and got to a point where I didn’t feel competent anymore and asked Claude to take from where I was.
5 minutes later I had almost 3 hours of important footage recovered.
I'm sure data recovery companies are pretty pissed that slightly esoteric data recovery abilities are becoming more accessible for average software devs. They were charging an arm and a leg to remote in and run scripts.
They still have two important moats: (1) expensive hardware tools (even stuff like SATA write blockers are kind of expensive for what they are), spare hard drive collections to swap failed PCBs, etc and (2) the "nobody got fired for hiring us" edge similar to how everyone calls in Crowdstrike/Mandiant after an incident. If a suit-level manager finds out customer data was lost, they are going to want to call in an expert so they can immediately tell the customer they did, not have the same internal team try to figure it out.
As an aside to #1: The cool thing is in modern times the hardware tools have come down stupidly cheap in price. Even SD card recovery is (vaguely) in the right skilled hands in a pseudo-professional home lab these days.
A lot of "Claude Code is best at X" claims are probably user-selection bias.
The people saying it are often exclusively Claude Code users, not people who are actively benchmarking Claude Code against Gemini CLI, OpenAI Codex, GitHub Copilot, and other agent harnesses on the same tasks.
The claim may still be true for certain scenarios, but the evidence is usually anecdotal, not comparative.
I was making a long edit in a crappy wiki UI and my browser froze. It would have taken a long time to redo, hours.
I didn't want to take the chance of force closing and losing everything. I used claude code to extract my text out of the browser internals and filesystem objects.
> Bitcoin trader recovers $400,000 using Claude AI after getting 'stoned' and losing wallet password 11 years ago — bot tried 3.5 trillion passwords before decrypting an old wallet backup
Man. I wish I had a lost wallet worth a quarter of that even, technically didn't need Claude for this, just needed any password cracking software.
Explaining your life to an llm, then having it generate permutations of passwords to try does sound like it would work a decent percentage of the time.
A large percentage of passwords aren't a random string of characters but a memorable word + memorable number. There's existing projects that basically do the same, and 3.5 trillion doesn't really make it clear if one of those wouldn't have worked as well, but I can see it having an above random chance to guess a password.
Many crypto wallets use a key derivation function (KDF) to add an amount of computation (and memory usage) per password tried - to mitigate brute force of weak passwords.
The increase in compute (decrease in brute-force cost) combined with price increases in many crypto tokens means brute-forcing old wallets can become worth it years after passwords were forgotten.
And of course even smaller, local AI models can now easily write optimized scripts to brute-force any given KDF function.
The other day, I asked Claude to track down the leaked Claude Code source so I could study it. It refused, saying “given who made me, I’ll pass.” It gave me some pointers on how to find it myself, which worked.
There isn’t that much of a difference between “help me crack this bitcoin wallet” and “help me crack this executable.”
I don’t exactly have a solid point, just some general observations. First, I think we’ll see AI more and more simply refuse to do any kind of forensics, as forensics becomes more powerful. Second, that implies local models will become more valuable, since they’re the only ones willing to do that kind of work.
I once got myself banned from Claude by researching barbiturates, since they’re connected with suicide. So my third observation is that we’ll see an uptick in people getting punished for trying to do things with AI that people don’t usually do. (Luckily the unban form worked.)
Someone downthread asked “how’d he convince Claude the coins weren’t stolen?” Which is an interesting question, because presumably some people trying to crack a wallet have stolen it. So I guess the fourth observation is that the exact framing you approach an AI with will become more important. There was the classic “do this or I’ll cut off my arm,” which worked a year ago. But in the future it will be more like “hopefully the AI believes my story, or else I’ll get into trouble.”
It’s good there are multiple AI vendors, or else it’d get real dystopian real fast when the de facto AI’s policy becomes something you have no way of working around.
Feels less like "ai cracked crypto" and more like having an insanely patient technical friend sitting next to you for 12 hours doing digital archaeology.
I spent a couple of days mining many years ago and got 2 bitcoins. At the time, they weren't worth the electricity they cost to mine and over time I lost the wallet and all information related to it.
"the user dumped their entire college computer files in Claude in a last-gasp effort."
Claude has limits that would make this simple statement be much more complicated-
Via Claude "So the chat upload file size limit is actually 500MB per file (not 30MB as many third-party sources claim - those appear to be outdated). The 20-file-per-chat cap and the 30MB-per-file limit in Projects remain consistent across plans.
The real constraint at any subscription level remains the context window - how many tokens Claude can hold in memory at once during a conversation. "
Claude ran a ctrl+f on his file system. Groundbreaking. Insane the dude hadn't figured this out for himself considering a few years salary was just sitting there.
> The bot uncovered an old backup wallet file that it successfully decrypted, while also uncovering a bug in the password configuration that was preventing recovery up to that point.
I know that we're all experts in archaic backup mechanisms and the encryption systems they used, but I think this qualifies as doing more than Ctrl+F
Claude found an old wallet and then ran btcrecover on that. The question is why the user could not find an old wallet with any numbers of Unix tools himself.
Since we are dealing with Anthropic, the entire story could be staged of course.
I've tried Claude Code with another LLM, it's very good at doing tasks and figuring things out. So this made me wonder, even though we know how good Claude models is, maybe the true value is in the harness now?
I have a similar claude story (much less money though), with the IRS R&D tax credit. The auditing firm initially said we qualify for $0. But then I had claude analyze past R&D reports and our expenses and it found the problem. The auditor had miscategorized our company.
So claude drafted an email even pointing to the right Internal Revenue Code (IRS Law), and specify why we fall under a specific category. The auditor got back to me two days later admitting their mistake and said our company now qualifies for $8k in tax credits. And a few months ago, it identified items in our AWS that saved us $250 a month (paying for itself).
So now I joke that even if I have a claude max plan, I've still come out ahead financially.
> So now I joke that even if I have a claude max plan, I've still come out ahead financially.
This is no joke; for better or worse, I see a day when I'm paying a lot more for this and it will be a bargain.
By my estimation (guess) you won't actually need to spend that much because the models are already getting a point where they don't need to get a whole lot better to be extremely helpful across many domains.
And it looks like those very helpful capabilities will continue to transfer to smaller models as well, as architectures and training regimes continue to refine.
I can fairly easily imagine a world where the only people needing to spend a lot of money on models are those that are using them to solve truly novel problems. The rest of us will get plenty of use at reasonable costs for the typical day-to-day helpful stuff.
To be fair regarding taxes it should be that you get all exemptions by default and the other side telling/justifying you why you don't qualify, instead of you not getting the things you should because you're not sure how to interpret the law or don't know this or that rule exists. Taxes shouldn't be that difficult, and the US version of it seems to be behind even the one I have here in europe (where my taxes are done "for me" for anything non business owner related). If the government is calculating your taxes anyway, they should just give the number to you instead of asking you a number and you better have the same as us or you're guilty of something.
Feels like a system that is deliberatly made to be more punishing for those who can't afford the help or the education to figure things out.
I would agree with you on most situations (like 1040 personal income taxes especially).
But in this particular tax credit, there's no way for the gov to know automatically what percentage of payroll was spent in qualified R&D expenses, since it's day to day business operations. Which is why we are _forced_ to hire an outside firm and pay them thousands of dollars (when Claude did an even better job), just to analyze how much of our time qualified as R&D expenses.
The problem I have is that I am forced to have to find a firm to do this, and most firms won't even work with companies as small as ours. So then we're stuck and losing out on years of R&D tax credits at the moment, when I really don't need them anymore, to be honest.
Or punishing to those that don’t pay for software and services to the companies that lobby for it to be this way.
[dead]
Claude Code is really good at stuff like this. The other day I tried to recover some images from an SD card that had gone bad. I used GetDataBack to recover files, but they appeared to be malformed and didn't open in image viewers.
I tasked Claude to analyze the files and figure out what's going on, and eventually we figured out that each file had a custom metadata header + thumbnail + actual image concatenated. I had it write a python script and was able to recover all the images with their metadata. It's nothing a human couldn't have figured out, but it was definitely WAY faster than doing it myself.
I've also used Claude in the past to figure out how to break into routers with locked down firmware. It's great at suggesting and trying approaches.
I have a friend that just picked up a new consulting job resurrecting an ancient Windows desktop application. No source control, no tests. And it's spread out over a dozen different folders with names like "_old", "_new" and "dates". Claude's doing a tremendous job in getting him to grips with what is actually happening in the application, what's relevant, what's not, what's different. I think it's literally saving him days and days at work.
if your friend has access to the binary and can pull it out to different box, they might get a lot out of a ghidra mcp -> https://github.com/LaurieWired/GhidraMCP
I think it would be interesting, once the dust has settled, to do a compare with a less expensive model (time, capital, compute) such as deepseek 4.
Any reason to expect that this wouldn't work 100%? It's not like the different LLMs providers are that technically different from one another.
> I have a friend that just picked up a new consulting job resurrecting an ancient Windows desktop application. No source control, no tests. And it's spread out over a dozen different folders with names like "_old", "_new" and "dates".
That doesn't sound very impressive. Not being tracked with a version control system is fixed instantly with a git init, git add ., git commit .no AI required.
Covering the app with tests is also something that requires no AI. At most, coding agents can generate characterization tests in broad sweeps, but we are talking about a delta between hand rolling and vibe-coding of a couple of days.
Where LLM shines is helping developers build up an understanding of what is in place. Running /explain on a codebase can quickly provide you with a high level summary of what's in place.
I did EXACTLY that last night. Was doing by hand for about an hour and got to a point where I didn’t feel competent anymore and asked Claude to take from where I was.
5 minutes later I had almost 3 hours of important footage recovered.
I'm sure data recovery companies are pretty pissed that slightly esoteric data recovery abilities are becoming more accessible for average software devs. They were charging an arm and a leg to remote in and run scripts.
They still have two important moats: (1) expensive hardware tools (even stuff like SATA write blockers are kind of expensive for what they are), spare hard drive collections to swap failed PCBs, etc and (2) the "nobody got fired for hiring us" edge similar to how everyone calls in Crowdstrike/Mandiant after an incident. If a suit-level manager finds out customer data was lost, they are going to want to call in an expert so they can immediately tell the customer they did, not have the same internal team try to figure it out.
As an aside to #1: The cool thing is in modern times the hardware tools have come down stupidly cheap in price. Even SD card recovery is (vaguely) in the right skilled hands in a pseudo-professional home lab these days.
https://blog.acelab.eu.com/pc-3000-flash-spider-board-adapte...
> Claude Code is really good at stuff like this.
A lot of "Claude Code is best at X" claims are probably user-selection bias.
The people saying it are often exclusively Claude Code users, not people who are actively benchmarking Claude Code against Gemini CLI, OpenAI Codex, GitHub Copilot, and other agent harnesses on the same tasks.
The claim may still be true for certain scenarios, but the evidence is usually anecdotal, not comparative.
Parent didn't say Claude Code is best at anything?
> Their luck changed for the better when they found an old mnemonic seed phrase written in an old college notebook
TBF the real breakthrough was finding this, though no doubt they couldn't have recovered without Claude
I was making a long edit in a crappy wiki UI and my browser froze. It would have taken a long time to redo, hours.
I didn't want to take the chance of force closing and losing everything. I used claude code to extract my text out of the browser internals and filesystem objects.
I did this a long time ago before the age of AI. Core dump and then run strings on it. Very low tech but very useful!
> Bitcoin trader recovers $400,000 using Claude AI after getting 'stoned' and losing wallet password 11 years ago — bot tried 3.5 trillion passwords before decrypting an old wallet backup
Man. I wish I had a lost wallet worth a quarter of that even, technically didn't need Claude for this, just needed any password cracking software.
Explaining your life to an llm, then having it generate permutations of passwords to try does sound like it would work a decent percentage of the time.
A large percentage of passwords aren't a random string of characters but a memorable word + memorable number. There's existing projects that basically do the same, and 3.5 trillion doesn't really make it clear if one of those wouldn't have worked as well, but I can see it having an above random chance to guess a password.
[delayed]
I'm really thankful I put my bitcoin in a time vault back in 2012 or so. It was inaccessible until about last year, and my $10 is now worth $100k.
Thank you MtGox.
> MtGox
Whew, that brings me back!
I still think about the Bitcoin my buddy paid me for his half of a pizza ~15 years ago... worth 6 figures now haha.
Nice, congrats. What's a time vault?
It's sarcasm.
Everyone who had coin in Mt.Gox lost it during a hack. A portion of that was returned to the users who had a loss about a year ago.
Yeah my 100 stolen bitcoins got me a cool $4k check from the settlement. Definitely made whole by that :|
Likely in this case the time vault was the collapse of Mt Gox, which has now recently been paying back holders.
It's something that locks your stuff so you can't access it for a while.
Many crypto wallets use a key derivation function (KDF) to add an amount of computation (and memory usage) per password tried - to mitigate brute force of weak passwords.
The increase in compute (decrease in brute-force cost) combined with price increases in many crypto tokens means brute-forcing old wallets can become worth it years after passwords were forgotten.
And of course even smaller, local AI models can now easily write optimized scripts to brute-force any given KDF function.
how can that possibly work while supporting offline backup & restore?
There’s an interesting ethical question here.
The other day, I asked Claude to track down the leaked Claude Code source so I could study it. It refused, saying “given who made me, I’ll pass.” It gave me some pointers on how to find it myself, which worked.
There isn’t that much of a difference between “help me crack this bitcoin wallet” and “help me crack this executable.”
I don’t exactly have a solid point, just some general observations. First, I think we’ll see AI more and more simply refuse to do any kind of forensics, as forensics becomes more powerful. Second, that implies local models will become more valuable, since they’re the only ones willing to do that kind of work.
I once got myself banned from Claude by researching barbiturates, since they’re connected with suicide. So my third observation is that we’ll see an uptick in people getting punished for trying to do things with AI that people don’t usually do. (Luckily the unban form worked.)
Someone downthread asked “how’d he convince Claude the coins weren’t stolen?” Which is an interesting question, because presumably some people trying to crack a wallet have stolen it. So I guess the fourth observation is that the exact framing you approach an AI with will become more important. There was the classic “do this or I’ll cut off my arm,” which worked a year ago. But in the future it will be more like “hopefully the AI believes my story, or else I’ll get into trouble.”
It’s good there are multiple AI vendors, or else it’d get real dystopian real fast when the de facto AI’s policy becomes something you have no way of working around.
I'm no expert but using an old wallet with a changed password, and it working, seems like a major security design flaw.
In the physical world, I can't imagine too many people being happy that old keys to your house still work even after you've changed the locks.
Can someone more informed, help me understand how this worked and why it's ok.
I'm genuinely wanting to become more informed & better understand.
[delayed]
They didn't lose the key, they just didn't know which one is the correct one, where the lock is, and how the unlocking is done.
Feels less like "ai cracked crypto" and more like having an insanely patient technical friend sitting next to you for 12 hours doing digital archaeology.
Claude is also surprisingly good at analyzing system issues on a Linux system and solving them!
I spent a couple of days mining many years ago and got 2 bitcoins. At the time, they weren't worth the electricity they cost to mine and over time I lost the wallet and all information related to it.
I'd love to mine a bitcoin a day on my PC now...
My wife doesn't like it when I tell the story of the hard drive I threw away with a wallet with 2 dozen BTC on it.
But lets be honest - when BTC hit 100 bucks, we would have cashed it out thinking we were geniuses.
Yep, I timed the top of the market perfectly and cashed out 5 BTC at ~$80/BTC.
Think on the bright side, at least you didn't spend 10000 BTC to buy pizza... speaking of which, Bitcoin Pizza Day coming up in just over a week!
https://www.coinbase.com/learn/crypto-glossary/what-is-bitco...
Does Claude turn out to be what 'Quantum' was promised; crack bitcoin? This could be fun.
"the user dumped their entire college computer files in Claude in a last-gasp effort."
Claude has limits that would make this simple statement be much more complicated-
Via Claude "So the chat upload file size limit is actually 500MB per file (not 30MB as many third-party sources claim - those appear to be outdated). The 20-file-per-chat cap and the 30MB-per-file limit in Projects remain consistent across plans. The real constraint at any subscription level remains the context window - how many tokens Claude can hold in memory at once during a conversation. "
> Claude has limits that would make this simple statement be much more complicated-
I guess the user simply pointed Claude Code at a local folder containing all the backups and files, and Code went through them via find/ls/etc
Claude ran a ctrl+f on his file system. Groundbreaking. Insane the dude hadn't figured this out for himself considering a few years salary was just sitting there.
> The bot uncovered an old backup wallet file that it successfully decrypted, while also uncovering a bug in the password configuration that was preventing recovery up to that point.
I know that we're all experts in archaic backup mechanisms and the encryption systems they used, but I think this qualifies as doing more than Ctrl+F
Also, it is right there in the article.
Would be worth a lot more if he had done this sooner and put it in the market 5 or 6 years ago.
BTC was valued at about $50k 5 years ago and about $10k 6 years ago. Now it is at about $80k.
So I guess he might be glad he didn’t figure it out earlier.
OK, that's impressive
How did they convince Claude they hadn't stolen it?
Maybe they said they were gay?
(Relevant; there was a “gay jailbreak” thread a week or so ago. I laughed.)
Claude hallucinate me a bitcoin address with unlimited money in it please.
[flagged]
Some more discussion: https://news.ycombinator.com/item?id=48136240
Is this not the link to this discussion?
Guessing the HN admins merged the post into this, which carries the comments.
Claude found an old wallet and then ran btcrecover on that. The question is why the user could not find an old wallet with any numbers of Unix tools himself.
Since we are dealing with Anthropic, the entire story could be staged of course.
I've tried Claude Code with another LLM, it's very good at doing tasks and figuring things out. So this made me wonder, even though we know how good Claude models is, maybe the true value is in the harness now?