A Ledger hardware wallet reseller was exposed for implanting chips to steal seed phrases, with over $90 million stolen, involving more than 300 wallets. In the AI era, the risk of single-device custody has surged. The article proposes five defense strategies: multisig to distribute private keys, AI-assisted custody solutions, domain verification, screenshot interpretation before signing, and security warning monitoring, teaching you to use AI to fight AI and protect your crypto assets.
Author: Lightning HSL
Something big has happened with hardware wallets again. A hardware wallet sold by a Ledger distributor had a chip implanted by someone, allowing coins to be stolen. Someone posted on X (note, this is not what the official statement said) saying that a circuit board was hidden under the Ledger screen, capable of stealing the mnemonic phrase when you write it down and sending it out.
Now it is said that the stolen amount exceeds 90 million U, involving more than 300 wallets.
A long time ago, I wrote an article saying I was not optimistic about hardware wallets, "Using a Dedicated iPhone + Wallet App Is Better Than Using a Hardware Wallet."
At that time, I thought: a dedicated iPhone with a wallet app installed is secure enough.
But now I think it is not enough. AI is too powerful. Attackers have more tools at their disposal: phishing pages look more convincing, fake customer service sounds more convincing, scams can be mass-customized one person at a time, and those 0day vulnerabilities we cannot understand can all have their threat amplified by AI. More attacks that ordinary people simply cannot figure out are being amplified by AI.
But we can also use AI to strengthen our defenses.
1. Use multisig, with keys stored on different devices and in different wallets
First, I still want to reject the view I expressed a few years ago that one device is enough. I think now we still need to use multiple devices for multisig to be secure enough. The risk of a single point being breached is still too great, even if you shut down your device every day and only turn it on when you use it.
For coins that need to be moved frequently, still use multisig. Buy one more device, manage a few more private keys, and you gain several more layers of security. At the very least, it should be 2-of-3.
Ethereum and other EVM chains have safe.global, which is very easy to use. One iPhone with a wallet app counts as one private key, one Mac with Chrome + an extension wallet counts as the second private key, and the third private key is generated on the iPhone, backed up, and then deleted.
In this way, unless your iPhone and Mac are compromised at the same time, you will be much safer.
For something like Bitcoin that does not need to be moved for years and does not need to go on any websites or do any DeFi, its only use is to deposit to and withdraw from exchanges. I feel it does not need multiple devices for multisig.
I feel this method is already secure enough: use a Bitcoin Core node wallet, buy a dedicated MacBook to install Bitcoin Core, and add a 2 TB hard drive to download the block data. The risk can already be reduced to a very, very low level, because Bitcoin is too simple and has very few attack points.
Although BTC also has multisig wallets, I always feel the branding is not strong enough, and Bitcoin script's native multisig solution requires too high a technical threshold. If you are not careful, your own coins could be treated as miner fees.
The important thing is to use a dedicated machine for a dedicated purpose. Do not use your daily phone or computer to manage coins. In particular, do not install AI.
Below I will still share what I think AI can help with in managing coins.
2. Let AI help you come up with a custody plan
Investing is personalized, and the plan for custodying coins is actually personalized too. I used to spend enormous effort writing many plans for managing wallets myself. Now that AI is here, these specific plans really feel worthless.
For custody plans, you no longer need to look for other articles. Just chat directly with AI.
Explain your situation clearly to AI: how many layers your assets are divided into, which are used often and which will not be touched for years; how often you move them; which chains you use; how complex an operation you can accept; and if something happens to you, how your family should take over.
Let it produce a plan. Then switch roles and have it play someone who wants to steal this money, telling you which link is easiest to attack. After revising, attack again, and go back and forth several rounds.
We trust AI too much and can fully authorize our computers to AI, so the computers and phones used for custodying coins must never have AI installed. This point really requires special attention. AI is too powerful, and we are too weak.
Also, is it safe to let AI know that I hold coins?
My view is that it is much safer than letting a distributor know when you buy a hardware wallet. The distributor knows your name, phone number, and address, and also handled the device on which your private key was born. Once you place an order, you cannot avoid these things.
Actually, the greatest security is still that no one knows you hold coins. Today that is impossible. Exchanges know, the store selling devices knows, and anyone can see on-chain addresses. The only choice is: who knows, and how much they know. Trusting AI is always better than trusting a specific individual.
3. Let AI check the source of URLs and apps
Dealing with coins always involves various websites, especially when doing DeFi, where there is always a bunch of websites to visit.
My previous rule was that for any website, I had to verify whether the domain was correct from three different information sources. Common information sources include whether official sites like ethereum.org/bitcoin.org and CMC have listed it; whether Google has checked it; whether x.com has checked it; and whether I have asked a truly trusted friend.
Now it is better. With AI, have you asked AI?
Checking domains is something AI is extremely good at, and using the free ones, Gemini, ChatGPT, Grok, and Doubao, and asking each once can serve as one of the important verification methods. It is best to use the paid ones and directly pull the reasoning intensity to the highest level when asking.
The same goes for apps. Before downloading a wallet app, send a screenshot of the App Store page to AI, and AI will check it for you very clearly. Chrome extension wallets should also be checked by AI.
The specific way to ask AI is also extremely simple. You just ask it whether it is legitimate. Now whenever I click on any new domain or download a new app, I ask AI once: is this thing legitimate?
4. Before signing, take a photo and send it to AI
This is the point I most want to share this time.
Actually, now for most ordinary transactions, such as transferring coins, authorizing, and interacting with mainstream contracts, MetaMask will display the specific meaning of the transaction in the signing interface. If you read carefully before signing, you can understand what will change in your wallet before and after this signature.
Fields like to, value, data, and nonce are now explained very well by wallets.
A few years ago, things were not this good. At that time, in the signing action interface of the wallet, all you could see was a string and a pile of messy characters like 00, fff, 1234.
Back then there was a website in 4byte.directory, and I often went there to paste transaction signature content to see what happened in the transaction.
If you still are not very clear about what transaction signing is, cannot understand these signature contents, and are not at ease with your transactions, then now there is AI, and AI can teach you to read these.
Take a photo or screenshot of the wallet pop-up interface before signing and send it to AI to read. Rest assured, this interface has no private key and no mnemonic phrase. The contents of this interface will be made public across the entire network after you click confirm. AI's interpretation will clearly tell you what will happen to your transaction after you sign. Then you will know whether you have been phished.
I think using a test wallet and having AI fully interpret every step of the interface in your operations is really very beneficial. Also, do not directly have AI read every screenshot of yours in real production. I am only talking about a test wallet, first using AI to learn to read the specific meanings of various wallet operations.
In real production, do not install AI on the phone and computer where the wallet is located. If you must have AI interpret your transaction signature content, please use the photo method.
When using the Safe wallet, the Safe wallet itself will explain a lot of the transaction content, but by the time it reaches the MetaMask signing stage, the transaction information becomes unclear again. If you are afraid, you can send that pile of 00, fff and other content to AI for interpretation. Once interpreted, you will know what will happen after your transaction is signed.
When using Safe, Safe wallet's own explanation of the transaction is not final. The explanation in the MetaMask signing interface is final. In the Bybit theft, the Safe explanation page was tampered with, so you must rely on the transaction field explanation in the MetaMask pop-up as the standard.
5. Let AI watch security warnings for you
There is another thing you can do with AI: let AI watch whether the hardware, software, contracts, and platforms you care about have vulnerability reports and security announcements.
Now Grok bot can perfectly monitor x.com. You can have Grok bot monitor the x.com accounts you care about. Give Grok bot a group of X accounts, such as MetaMask, SlowMist, Lido, ... and let Grok bot watch whether they have posted anything security-related, and email you if they have.Also, if you really receive a warning, verify it yourself through official channels, and don't click links in emails or posts. Be careful not to get phished. Of course, sending an AI to verify is also fine.
When you make a major financial transfer, read this email first.
The End
I think AI will ultimately be a force for good, making our online environment safer.