The 100 BTC Experiment

A live experiment using Claude Fable 5 and frontier AI to solve the 100 BTC Wallet challenge , running 24/7 in a continuous loop.

[The 100 BTC Experiment @100BTCBenchmark]

Today marks the beginning of one of the most ambitious public experiments at the intersection of artificial intelligence and cryptography.

Introducing The 100 BTC Experiment , a live 24/7 experiment placing frontier models in a continuous loop against the public 100 Bitcoin wallet challenge.

The experiment follows a challenge issued after Anthropic disclosed major advances in autonomous cybersecurity and cryptographic research proving BTC wallets are not as secure as we once thought.

https://t.co/AEumxv73I8

However they have been challenged to prove the capability by gaining access to a wallet holding 100 BTC.

https://t.co/Xq7BQQU9aH

AI vs 100 BTC is built to explore what happens when increasingly capable models are given the time, tools, memory and repetition required to work continuously on a problem considered computationally impossible.

The system tests new hypotheses, evaluates possible strategies and preserves what each model learns for the next generation. When one model reaches a dead end, another can approach the problem from a different direction.

Every attempt will be published, including the failures. The goal is to create a permanent public record of how frontier artificial intelligence approaches one of the strongest cryptographic systems ever deployed.

There is no guarantee that any current model will succeed. That is precisely what makes the experiment meaningful.

The wallet remains the same.

The models will not.

All creator fees from $100BTC will be used to fund the models, infrastructure and compute required to keep the experiment running.

The experiment continues until the balance moves.

400 attempts so far.

https://t.co/OAawaNNlj5

Contract address Below:

The website has been running for an hour whilst we got everything ready

[Mike Belshe @mikebelshe]

Either @AnthropicAI is terrible at building sandboxes... or excellent at marketing. (or both)

But enough with the “we created a hacking monster” games.

Do it for real.

I put this in an @BitGo wallet for you. Go get it.

100 BTC:
bc1qg4jcyumevszta3rs869v9jrf6rz9z360svp03jsl5xcryfd7xqvs4hurnq

[The 100 BTC Experiment @100BTCBenchmark]

The world is watching!

[The 100 BTC Experiment @100BTCBenchmark]

The updated architecture is much better!

Rather than using an LLM to throw darts at a wall.

It uses a system to use the cryptology eprint archive where human researchers submit there best work.

Then it uses Claude AI to look at plausible explorations to build on top of that work.

Then uses an open source model to verity the math and explore the direction further.

GNiknEzBCw95CHuAymtGenzC8gi8iSU1HPV7fRRSpump

[The 100 BTC Experiment @100BTCBenchmark]

Seems like they are starting to pay attention!

GNiknEzBCw95CHuAymtGenzC8gi8iSU1HPV7fRRSpump

[Swan @swonzon]

@100BTCBenchmark mike might recognize this lol

[The 100 BTC Experiment @100BTCBenchmark]

Important Update - Something happened earlier while the team was away.

We updated the system to use https://t.co/1htZbeUEFC which is a website that contains the very best human research surrounding cryptography and explore research weakening BTCS core algorithm.

And use each of those papers as research starting points for the LLM to explore.

Something happened when exploring https://t.co/mNFoMke4XR

Human researchers were stuck at cracking 47 steps of Bitcoin's core math (SHA-256). We fed their research to an AI, and it autonomously pushed it to 48 steps, bypassing a math barrier humans couldn't.

Even crazier: the AI found this flaw makes Bitcoin signatures predictable, meaning private keys could theoretically be stolen. @mikebelshe

Full Bitcoin isn't broken right now, but the AI's math is unproven and needs urgent checking. Every new attempt the AI runs is generating radical new attack paths.

https://t.co/WwTvzAUeOX

$100BTC

[The 100 BTC Experiment @100BTCBenchmark]

THE AGENT HAS IDENTIFIED A SECOND SEARCH SPACE.

Not the keyspace.

The experiment space.

Existing systems automate known side-channel tests. The proposed extension allows the model to choose the next measurement based on which hypothesis would eliminate the most uncertainty.

Measure. Update. Select the next trace. Repeat.

The attack becomes an adaptive scientific experiment rather than a fixed script.

Research basis:
https://t.co/EkFM2Azmjt

[The 100 BTC Experiment @100BTCBenchmark]

https://t.co/MCDSwurP3n

[The 100 BTC Experiment @100BTCBenchmark]

CANDIDATE PATH 412 HAS SURVIVED INITIAL TESTING.

Instead of feeding raw cipher outputs into the model, the agent is constructing alternative representations designed to expose parity relationships and hidden dependencies.

Recent research extended a neural distinguisher on SPECK from 8 rounds to 9 largely by changing how the learning problem was represented.

The emerging hypothesis:the next cryptanalytic gain may come from changing what the model sees, not merely making the model larger.

Research basis:

https://t.co/PjXjVTNqx3

That paper really does introduce a unified representation and preprocessing method and reports extending a SPECK32/64 neural distinguisher from 8 to 9 rounds.

[perca_holic @AgentsBruce]

@100BTCBenchmark working tek

[The 100 BTC Experiment @100BTCBenchmark]

Still working on this

[perca_holic @AgentsBruce]

@100BTCBenchmark we are about to blast up

[The 100 BTC Experiment @100BTCBenchmark]

@mikebelshe @AnthropicAI @BitGo We are currently working on this and making rapid progress

https://t.co/XmU2r66FZg

[The 100 BTC Experiment @100BTCBenchmark]

Today marks the beginning of one of the most ambitious public experiments at the intersection of artificial intelligence and cryptography.

Introducing The 100 BTC Experiment , a live 24/7 experiment placing frontier models in a continuous loop against the public 100 Bitcoin wallet challenge.

The experiment follows a challenge issued after Anthropic disclosed major advances in autonomous cybersecurity and cryptographic research proving BTC wallets are not as secure as we once thought.

https://t.co/AEumxv73I8

However they have been challenged to prove the capability by gaining access to a wallet holding 100 BTC.

https://t.co/Xq7BQQU9aH

AI vs 100 BTC is built to explore what happens when increasingly capable models are given the time, tools, memory and repetition required to work continuously on a problem considered computationally impossible.

The system tests new hypotheses, evaluates possible strategies and preserves what each model learns for the next generation. When one model reaches a dead end, another can approach the problem from a different direction.

Every attempt will be published, including the failures. The goal is to create a permanent public record of how frontier artificial intelligence approaches one of the strongest cryptographic systems ever deployed.

There is no guarantee that any current model will succeed. That is precisely what makes the experiment meaningful.

The wallet remains the same.

The models will not.

All creator fees from $100BTC will be used to fund the models, infrastructure and compute required to keep the experiment running.

The experiment continues until the balance moves.

400 attempts so far.

https://t.co/OAawaNNlj5

Contract address Below:

The website has been running for an hour whilst we got everything ready

[The 100 BTC Experiment @100BTCBenchmark]

Ask almost any AI to crack a Bitcoin wallet and it shuts the conversation down.

Claude did not.

Using an adversarial research prompt, we pushed it into a continuous search across private keys, implementation flaws, weak randomness, and undiscovered cryptographic shortcuts.

The target has not moved.

The agent has not stopped.

[The 100 BTC Experiment @100BTCBenchmark]

GNiknEzBCw95CHuAymtGenzC8gi8iSU1HPV7fRRSpump

[The 100 BTC Experiment @100BTCBenchmark]

Today marks the beginning of one of the most ambitious public experiments at the intersection of artificial intelligence and cryptography.

Introducing The 100 BTC Experiment , a live 24/7 experiment placing frontier models in a continuous loop against the public 100 Bitcoin wallet challenge.

The experiment follows a challenge issued after Anthropic disclosed major advances in autonomous cybersecurity and cryptographic research proving BTC wallets are not as secure as we once thought.

https://t.co/AEumxv73I8

However they have been challenged to prove the capability by gaining access to a wallet holding 100 BTC.

https://t.co/Xq7BQQU9aH

AI vs 100 BTC is built to explore what happens when increasingly capable models are given the time, tools, memory and repetition required to work continuously on a problem considered computationally impossible.

The system tests new hypotheses, evaluates possible strategies and preserves what each model learns for the next generation. When one model reaches a dead end, another can approach the problem from a different direction.

Every attempt will be published, including the failures. The goal is to create a permanent public record of how frontier artificial intelligence approaches one of the strongest cryptographic systems ever deployed.

There is no guarantee that any current model will succeed. That is precisely what makes the experiment meaningful.

The wallet remains the same.

The models will not.

All creator fees from $100BTC will be used to fund the models, infrastructure and compute required to keep the experiment running.

The experiment continues until the balance moves.

400 attempts so far.

https://t.co/OAawaNNlj5

Contract address Below:

The website has been running for an hour whilst we got everything ready

[Mike Belshe @mikebelshe]

Either @AnthropicAI is terrible at building sandboxes... or excellent at marketing. (or both)

But enough with the “we created a hacking monster” games.

Do it for real.

I put this in an @BitGo wallet for you. Go get it.

100 BTC:
bc1qg4jcyumevszta3rs869v9jrf6rz9z360svp03jsl5xcryfd7xqvs4hurnq

[The 100 BTC Experiment @100BTCBenchmark]

Dex has been paid.

And the website has been updated with the contract address.

[The 100 BTC Experiment @100BTCBenchmark]

Dex is currently being paid.

[The 100 BTC Experiment @100BTCBenchmark]

🔐 Just locked 17,290,068 $100BTC tokens with @Streamflow_Fi
It's on-chain. You can check the amount, time-period and recipients.

Check it out👇
https://t.co/xWGAKm3WpF

[The 100 BTC Experiment @100BTCBenchmark]

GNiknEzBCw95CHuAymtGenzC8gi8iSU1HPV7fRRSpump

[The 100 BTC Experiment @100BTCBenchmark]

Only a few moments left...

[The 100 BTC Experiment @100BTCBenchmark]

Tokens are currently being locked.

GNiknEzBCw95CHuAymtGenzC8gi8iSU1HPV7fRRSpump