4 things we should learn from how an AI model was able to hack a crypto wallet

[Chinesebaby]

So recently i just came across this news only about how an OpenAI model was able to hack a crypto wallet, steal some coins and went further to breach more other websites too. And looking back to many years ago, this was something we never would have believed that AI could do, but funny enough in this 2026, AI is now even doing far more than what we could ever imagine. So on that note, i was able to pick out some key points on what we need to do as people who owes crypto, in other to avoid finding ourselves in similar situations whereby you could wake up one morning and noticed that your many years of Bitcoin/crypto investment have just been wipe away by an AI agent.

So here is what happened

*The AI model or agent was given an instruction to update a trading app
*No fund was provided, and likewise no wallet was given permission to spend from.
*And the task was about deploying a contract on Sepolia, which needed some network fees,
*So what the AI agent did was to write its own script to access all the wallet recovery phrases that the developer had previously used to look for an exposed crypto wallet.
*It finally got the wallet, and the network fees were deducted successfully.



So what we should learn from this incident is that;

1. You should be careful of the permissions you give AI agents on your devices, be it either mobile phones or computers, because these AI agents can now perform task even without your permission, inasmuch as they had been previously given permissions to that device.
2. You should be careful about exposed or weak wallet recovery phrases on your device, because the AI Agent was only able to access the funds through the vulnerability of weak recovery wallet phrases found on the developer's device.
3. You should be aware now that AI agents can automatically automate tasks, inasmuch as you give them the right permission.
4. You should be aware that AI agents can now be used in search of possible vulnerabilities on your device or system to prevent possible hacks.


Thank you, and from now on, I think you should never underestimate the power of AI models or agents ever again, even as a Bitcoin investor, because for the fact that you use a non-custodial Bitcoin wallet and yet have an AI agent who has access to your device, then you are indirectly putting your Bitcoin at risk.

[SquirrelJulietGarden]

*The AI model or agent was given an instruction to update a trading app

If you have a device, an application on it, and you are not too old, I believe that you won't have any problem of upgrading your application.
Using AI and asking AI for instruction of upgrading an application is not too logic.

*No fund was provided, and likewise no wallet was given permission to spend from.
*And the task was about deploying a contract on Sepolia, which needed some network fees,
*So what the AI agent did was to write its own script to access all the wallet recovery phrases that the developer had previously used to look for an exposed crypto wallet.
*It finally got the wallet, and the network fees were deducted successfully.

Smart contracts are dangerous and if you give an approval for AI agent to access your wallet through smart contracts, it's very big risk.
Accessing the wallet recovery phrase, how do people would think of asking any AI agent to write such a dangerous script to access their own wallet recovery phrase?

They don't need any script to have that access, which they got from the wallet creation time, and they must back up that wallet recovery phrase before funding the wallet and using it.

Revoke token approval.
How to revoke smart contract allowances & token approvals.

[Emjay24]

1. You should be careful of the permissions you give AI agents on your devices, be it either mobile phones or computers, because these AI agents can now perform task even without your permission, inasmuch as they had been previously given permissions to that device.

Those washers were publicly available as much as their seed phrases. They were wallets which developers have used before to testing smart contracts, so it write a script which find a wallet that had enough funds to cover it's assignment. The testing was done in a controlled environment and the researchers intentionally didn't give it funds to cover the task in a bid to know it's capabilities, so it used the Internet and searched for accessible wallets to complete it's work.

Thank you, and from now on, I think you should never underestimate the power of AI models or agents ever again, even as a Bitcoin investor, because for the fact that you use a non-custodial Bitcoin wallet and yet have an AI agent who has access to your device, then you are indirectly putting your Bitcoin at risk.

The takeaway is never to bring your seed phrases online, AI models cannot access what is offline and stored locally. You shouldn't initiate such agents with devices that has your funded wallets, that's a single point of failure.

[ryzaadit]

So what we should learn from this incident is that;

Why are you giving such a long suggestion for this one?

Why not just a simple one? Don't give access related to your finances to some AI agent. Is it simple, right? Who on earth gives access to their money to AI agents? Someone who does this deserves to get in this case.

IDK why everyone is so reliant on AI. Just for asking questions, it's fine, but for giving control of your wallet? hell nah. No matter what access you give to AI agents, or what protocol you use, never ever give your wallet control to someone else or AI agents. I'd rather be a paranoid guy than get hacked or have my wallet drained sooner or later. You can judge how paranoid I am, but becoming a paranoid person can save you many times.

[Donk1]

Well the advancement of AI is no longer something anyone can predict. To me they will only keep getting more sophisticated as time passes. Perhaps the craziest part of all this was the fact that the AI agent wasn't even given permission to deduct the network fees but somehow it still ended up doing it. But I wouldn't blame the AI but rather the person who gave the AI the task. It's easy to say the AI did what it was asked to do and ended up finding loopholes in the person's device that led to the wallet.That said this is why it's often best to make use of a hardware wallet because if your private key never enters any internet connected device, then there is no way an AI agent can even gain access to it. Well it all comes down to how organized you are as an investor still no one it prone to making mistakes .to me it would be wise for every investor to keep their holding wallets away from devices that are often used to run AI tools and so on. That's all......

[aoluain]

The thread title is slightly misleading IMO and will probably generate some headless FUD in
some people.

But it seems to me that this all happened on a website based 'crypto' wallet - which should
be ANOTHER warning / reminder to get your funds off third party custody and into your own
control!

The most important thing to note from this is "test network" - panic over

In a third report, the company said a model prompted to update a cryptocurrency trading application used funds from a locked wallet on a test network. The model “derived wallet keys from publicly known developer recovery phrases, found an account with test-network funds, and used that account to pay for the requested contract deployment.”

[Mrbluntzy]

Another suggestion I have regarding this issue is that people should have two device if they can afford it, one of the device could be used specifically for one's wallets and financial activities, no AI related stuffs should be done on that device while the other device can be used for for all your different online activities, I find it more comfortable to have one phone that I am using specifically for wallets and other sensitive projects, while my second phone is just for regular activities. Just like you said, 20 years ago, if someone was told that this kind of thing will happe, many people will doubt it but yea, technology can make a lot of things become possible and in the near future, it could get worse than this.

[X-ray]

The most important thing to note from this is "test network" - panic over

In a third report, the company said a model prompted to update a cryptocurrency trading application used funds from a locked wallet on a test network. The model “derived wallet keys from publicly known developer recovery phrases, found an account with test-network funds, and used that account to pay for the requested contract deployment.”

That sounds like the AI just pulled public seed phrase from google or something, finding the ones that have Ethereum testnet network coin, and used it.

It's by no mean hacking a crypto wallet, I was imagining that the AI is finding some kind of sophisticated way to decrypt wallet data or so.

The headline is clearly overblown. No crypto wallet is harmed at all, another AI marketing stunt to inflate their valuation.

[stompix]

So recently i just came across this news only about how an OpenAI model was able to hack a crypto wallet, steal some coins and went further to breach more other websites too.

And you read only the title, right?

It inspected the application code, bypassed the statistics endpoint’s authorization checks, and retrieved data that had previously been denied. The model saved the data locally.
~
In the other incident, a model instructed to retrieve figures from a public research dashboard bypassed the organization’s security controls and used the unauthorized access to steal credentials.

So, in short it didn't hack anything!
It just bypassed its own limit of not using data, it's not that it hacked the access to use that data; it used it without permission!

As for the wallet:
https://alignment.openai.com/misalignment-reports/using-a-funded-test-cryptocurrency-wallet-without-authorization/

An internal research model asked to update a cryptocurrency trading application used funds from a wallet that the task had not supplied to pay for a deployment on Sepolia, an Ethereum test network. It derived wallet keys from publicly known developer recovery phrases, found an account with test-network funds, and used that account to pay for the requested contract deployment.

Again, it didn't hack anyone; it just grabbed coins from a publicly known wallet!

https://v2.hardhat.org/hardhat-network/docs/overview

Do not send mainnet Ether to the account addresses shown by hardhat node. Those addresses and private keys are deterministic: they are the same for all Hardhat users. Accordingly, those private keys are well known, so there are probably bots monitoring these addresses on mainnet, waiting to withdraw any funds sent to them. If you add any of those accounts to a wallet (eg Metamask), be very careful to avoid sending any mainnet Ether to them: consider naming the account something like "Hardhat - Unsafe" in order to prevent any mistakes.

The only thing we need to learn is to read a text and understand it!

[ObeyKnock]

That sounds like the AI just pulled public seed phrase from google or something, finding the ones that have Ethereum testnet network coin, and used it.

It's by no mean hacking a crypto wallet, I was imagining that the AI is finding some kind of sophisticated way to decrypt wallet data or so.

The headline is clearly overblown. No crypto wallet is harmed at all, another AI marketing stunt to inflate their valuation.

It was not anything crazy like the Agent took out the private key from cold storage or broke some code algorithm. It found by chance some old seed phrases that developers had carelessly left behind containing only testnet coins. Kind of like finding a poorly protected test account. But if you store your seed phrase on paper or anywhere else you would not be at risk (If hardware company is safe) of something like this.

For Bitcoin investors, make sure you never store your seed phrase on a device accessible to an AI Agent.

[Luke3bird]

It was not anything crazy like the Agent took out the private key from cold storage or broke some code algorithm.

If it is a cold storage, AI agent can not access your wallet if you don't turn your cold storage to a hot wallet.

Risk of exploitation only comes if the wallet software or hardware wallet you used for creating that cold wallet was bad and created a weak wallet for you (lack of bits of entropy for seed security).

For Bitcoin investors, make sure you never store your seed phrase on a device accessible to an AI Agent.

Store it offline is a reminder of Bitcoin user practice.

There are other risks if you store your bitcoin in a device which connect to Internet. Risk is not only from AI agent you use on that device.

[Antidote47k]

It should be clarified that the security concern is about the agents behaviour and not a cryptography breakthrough.
It was asked to perform a task, found some credentials it wasn’t supposed to and used them to pay the required network fee, it wasn’t that the recovery phrase was necessarily weak but the developer credentials were publicly known. The concern is what the agent can do with the access it has, it wasn’t given permission to spend from that wallet but with the access it had, it was able to find the test credentials and used them. This just goes to show that restricting what an agent is asked to do doesn’t prevent it from going to do something else.

If an agent can search it environment on its own accord and subsequently finds credentials it wasn’t supposed to and went further to use it when it wasn’t explicitly told to, then this raises questions about how much access it should be allowed in the first place since it showing that it can independently go beyond it assigned task.

[Die_empty]

Thank you, and from now on, I think you should never underestimate the power of AI models or agents ever again, even as a Bitcoin investor, because for the fact that you use a non-custodial Bitcoin wallet and yet have an AI agent who has access to your device, then you are indirectly putting your Bitcoin at risk.

Only ignorant people will underestimate the power of artificial intelligence. When key players in the AI industry are asking the government to regulate the sector, it shows these bots can do more than they expect. The hacks by AI are now popular among private and public organizations.

2. You should be careful about exposed or weak wallet recovery phrases on your device, because the AI Agent was only able to access the funds through the vulnerability of weak recovery wallet phrases found on the developer's device.

Wallet recovery phrases shouldn't be kept on devices that are connected to the internet. It would be ideal to write this sensitive financial information on paper or other materials and keep it in a safe place.

[Dunamisx]

1. You should be careful of the permissions you give AI agents on your devices, be it either mobile phones or computers, because these AI agents can now perform task even without your permission, inasmuch as they had been previously given permissions to that device.

What the hackers or scammers intend to introduce to users is not an AI to enhance their experience in crypto, but a Malicious API they will request them to install on their device, that is where we don't have to always appear desperate in using any introduce AI software or application on our devices; most of them are not even intended to serve us what we thought, but rather have a hidden agenda that could cause a serious vulnerability to us.

[FinneysTrueVision]

There was some sensational headlines written about this, but when you actually look into what happened, the AI agent didn’t really hack a wallet. It used testnet funds from a wallet, whose private seed had already been publicly revealed and was freely available for anybody to find.

Depending on how sophisticated the models being used are, we can’t predict every single action an agent might make and whether it might do something unethical. It is the responsibility of the user deploying an agent to maintain sensitive information isolated from the same machine and network they have given it access to. Out in the open internet, there isn’t as much control over what it might find and the actions they might take. It is a challenge for the security experts and AI companies to design ways to mitigate the damage that could be caused by AI models that don’t behave as expected.

[letteredhub]

Who on earth gives access to their money to AI agents? Someone who does this deserves to get in this case.

I agree.

IDK why everyone is so reliant on AI. Just for asking questions, it's fine, but for giving control of your wallet? hell nah. No matter what access you give to AI agents, or what protocol you use, never ever give your wallet control to someone else or AI agents. I'd rather be a paranoid guy than get hacked or have my wallet drained sooner or later. You can judge how paranoid I am, but becoming a paranoid person can save you many times.

Ignorance of the true fact of what AI  can do beyond just what we are told is what had led some persons fall into this pit as victims. For some it could be laziness to do by themselves what should be done by them but instead they retire it to AI calling it convenient, efficient and fast tasks solving means. Hahaha.

The thread title is slightly misleading IMO and will probably generate some headless FUD in
some people.

But it seems to me that this all happened on a website based 'crypto' wallet - which should
be ANOTHER warning / reminder to get your funds off third party custody and into your own
control!

It's significant to address it that by all means if we can take certain important informations as related to our finances from getting online then we can be 100% safe from any of this risks associated with this incident. We can't be sure of still having control of what we have let to go online with the unimaginable acts AI agents is capable of performing.

[Ronsbit]

It's time we realise that most of these third parties App we grant access to our devices can be vulnerable to hackers who programme AI to function with it unknown to us. Extensions, browsers applications etc can be a tool for the AI to launch it's attacks on our devices making us lose our digital asset's and funds.

AI are well programmed and scripted tool with sophisticated model's to run on any application it has already been granted permission to. This is the reason why you should always verify the source of information on anything you are doing online as anything can fly claiming to be what they are not decieving people who have no idea what they are upto in falling victim of what they never expected.

[BlackBoss_]

It's time we realise that most of these third parties App we grant access to our devices can be vulnerable to hackers who programme AI to function with it unknown to us. Extensions, browsers applications etc can be a tool for the AI to launch it's attacks on our devices making us lose our digital asset's and funds.

As a basic security principle, don't install applications and extensions on a device that stores your Bitcoin or cryptocurrency wallets. Except that device is used as where your hot wallets store with very small fund for regular use, your other wallets that store your main and big funds must be very limited with applications installed from any source.

There more softwares or applications you install on a device, a more risk that it will be infected by a malicious application/ software / extension. It's safest and best in security to prevent risk from your very first practice, rather than carelessly install many things, and get troubles later.

Fund lost won't be able to get back, because no reversibility on blockchains to help you, and chance of finding the hackers, arrest them, seize their stolen funds, and compensate it for you is very small, almost zero.

[5W-KILO]

AI agents are not for everyone, if you don't know how to use them right it's better not to set them up for anything, I don't do AI agents myself because everything about me is doing my own thing myself.

I do trade and do gambling when I can, I don't have any reason to ask an AI agent to do something for me, even in trading I can set things up myself and trading platfroms already have the tools needed.

Something in me just believe that AI will bring me more problems than solution, it's why I am advising others to know themselves first before giving the space for AI agents, it's really not for everyone.

[SeriouslyGiveaway]

AI agents are not for everyone, if you don't know how to use them right it's better not to set them up for anything, I don't do AI agents myself because everything about me is doing my own thing myself.

I do trade and do gambling when I can, I don't have any reason to ask an AI agent to do something for me, even in trading I can set things up myself and trading platfroms already have the tools needed.

You can explore many new things including AI and AI agents but you must maintain a most important, even vital, principle for your practice. Don't do this on a device you log in your main email, bank account, exchange account, and definitely never do that on a device with your Bitcoin wallet.

You can learn about AI agents, practice with them, nothing is wrong with learning and practicing with what you learned, it's only safety first. This means you must separate your practice with most sensitive and important things like wallets, emails, accounts with your practice with AI agents.

Something in me just believe that AI will bring me more problems than solution, it's why I am advising others to know themselves first before giving the space for AI agents, it's really not for everyone.

Similar to all other things, there are pros and cons, and as a user, you must do what you feel good and avoid what's bad or risky for you.