Bitcoin Wallet Inaccessible for Over a Decade: Claude Succeeds Where Eight Weeks of Brute-Force Failed
Since 2014, when a bitcoin was worth about 250 dollars, an user known as @cprkrn on X had held 5 BTC in a wallet that became inaccessible over time. In May 2026, after more than eleven years, those cryptocurrencies were recovered with the help of Claude, the LLM model developed by Anthropic. The news reached over six million views on X, sparking a debate about what large language models can actually do in the context of applied cryptography.
What Happened to the Wallet
The wallet in question was a legacy P2PKH address, the standard format used before 2015. @cprkrn had a mnemonic phrase recovered later, but the current wallet file had been updated with a different password, set at a time the user could not remember precisely. For eight weeks, he conducted brute-force attacks via btcrecover, an open-source tool for recovering Bitcoin wallets, using GPUs rented through Vast.ai, testing about 3.5 trillion combinations. The total cost for renting the computing capabilities amounted to around 15 dollars, for an entirely failed attempt.
As a last resort, @cprkrn uploaded the entire content of his old university computer into Claude. The assistant located a wallet backup file dating back to December 2019, prior to the password change. We are talking about an operation that, at its essence, is forensic analysis on file systems: Claude identified the correct backup among a mass of unstructured data, recognized the wallet format, and linked that file to the mnemonic phrase the user already possessed. The password was noted in a notebook; once associated with the previous backup, it correctly decrypted the private keys.
HOLY FUCKING SHIT OMG CLAUDE JUST CRACKED THIS SHIT, THANK YOU @AnthropicAI THANK YOU @DarioAmodei NAMING MY KID AFTER YOU 😍 https://t.co/gObNirRDpS https://t.co/ByTdIM4d20 pic.twitter.com/xB5LUJb6Pe — 🍜 (@cprkrn) May 13, 2026
It must be emphasized that the central technical point does not concern the encryption of the Bitcoin protocol itself, which remains intact. The problem was at a higher level: btcrecover, in the implementation used, concatenated the sharedKey and password in the wrong order during decryption key derivation. Claude identified the bug in the tool's behavior, corrected the decryption logic, extracted the private keys in Wallet Import Format (WIF), and validated them on-chain against the target address. The textual output of the model, immortalized in a now-viral screenshot, stated: "PRIVATE KEYS DECRYPTED! WE GOT IT!!! THE 5 BTC IS YOURS!"
No, Claude Did Not "Crack" Any Wallet
The narrative spread on X has somewhat distorted the nature of the operation. Several posts spoke of Claude "cracking" a wallet or "breaking Bitcoin encryption," formulations that security experts promptly contested. In short, what Claude did was apply contextual reasoning to files and code: nothing comparable to an elliptic curve attack on which the security of Bitcoin private keys relies, a computationally unsolvable problem with current hardware, which researchers place in a timeline linked to quantum computers, at least five to ten years in the future according to prevailing estimates. We are faced with a case where the utility of AI lies not in brute-force cryptography, but in the ability to navigate complex, legacy, and fragmented technical environments that would require hours of manual analysis by an expert.
It is worth noting that Claude's contribution was twofold, involving research and debugging. The research, namely identifying the correct file among years of accumulated data, is the most immediately replicable part for anyone in a similar situation. The debugging of btcrecover, correcting the concatenation order of sharedKey+password, is the more technical component that actually unlocked the decryption. Professional wallet recovery services typically charge fees of 20% on the recovered value; in this case, the cost would have been around 80,000 dollars. @cprkrn achieved the same outcome in a single session with an LLM.
Implications for Dormant Bitcoins
According to estimates by Ledger, between 2.3 and 3.7 million BTC could be permanently inaccessible, with some estimates bringing the figure closer to 4 million coins. The reasons are the same as always: forgotten passwords, destroyed hard drives, recovery phrases never stored correctly, wallets abandoned in the early years of adoption when private key management was not part of any structured onboarding process. With bitcoin recovering at around 80,000 dollars, a laptop or a USB stick forgotten in a drawer can contain six-figure capital.
We cannot help but consider that this case opens a concrete scenario for those holding hardware from the 2010-2015 era: current LLM models can perform forensic analysis on file systems in a way accessible to non-technical users, lowering the entry barrier compared to traditional recovery tools that require specific expertise on wallet formats, key derivation scripts, and execution environment setup. On one hand, this makes the recovery of forgotten wallets more feasible; on the other, the same contextual analysis capability on encrypted files brings back to the forefront of the debate the secure management of backups and the separation between sensitive data and environments shared with cloud or AI services. However, what an LLM still cannot do is search hard disks in a landfill...