When Money Moves at Machine Speed, Security Has to Live in Hardware

Before You Dive In:

• Tokenized money moves the moment a key signs. The days of delay that let banks catch a bad payment are gone, so security has to do that job.
• AI makes attacks cheaper and agents faster, and anything in software can be attacked. The approval logic belongs in hardware.
• Quantum has no date, but upgrading the cryptography in chips, wallets, and keys takes years, so the work has to start before the threat arrives.

In August, I sat down with Ronit Ghose and Gaurav Dhar on the FUMO: Future of Money Podcast to talk about what happens when money moves at machine speed. We covered banks, stablecoins, quantum, and why I think security has to live in hardware. Two stories frame my answer.

In February 2016, hackers tried to steal nearly $1 billion from Bangladesh Bank’s account at the New York Fed. The heist was broken by two basic security checks: automated filters paused most of the transfers when a street address matched a blacklisted ship, and a bank clerk stopped another after spotting a single typo (“Fandation”). Out of $951 million targeted, only $81 million slipped through.

In April 2026, attackers drained about $280 million from Drift, a decentralized trading protocol. The drain reportedly took 12 minutes. Drift later said the operation followed months of preparation, so the drain was the fast part. North Korea’s Lazarus Group is widely blamed for the first heist and is believed to be behind the second.

Ten years apart, the two heists show what tokenization takes away: the pause in which a person can still say ‘stop’. I know that pause from inside a bank. I spent two decades at Morgan Stanley, Goldman Sachs, Citi, and Deutsche Bank before joining Circle in 2021 and then Ledger. My path to crypto was gradual: first an odd new word, blockchain, then stablecoins as a use case that could go mainstream. Tokenization is what led me to Ledger.

Banks Will Absorb Crypto Rather Than Lose to It

Crypto was easy to dismiss in its early years. Regulators finally noticed when it grew into a multi-trillion-dollar financial system with almost no oversight. In 2023, three US banks (Silvergate Bank, Signature Bank, and Silicon Valley Bank) failed in a single week and the failures of two of them were related to their exposure to the crypto ecosystem. This was a wake up call for banking supervisors who moved to keep crypto away from the banking system (and no doubt overreacted). That stance did not last.

Banks follow their supervisors, and the regulatory deep freeze began to thaw with the change of administration in the US and the passage of the Genius Act to regulate stablecoins in the US and earlier adoption of MiCa in the EU. Now, banks, exchanges, and payment companies are all moving into blockchain because it is a fundamentally more efficient way to record and transact value. The next phase of growth will run the other way: mainstream finance is coming to blockchain and given the size of the financial system (JPMorgan Chase’s balance sheet is just under twice the size of the entire crypto ecosystem), it is “TradFi” that will drive the growth of blockchain in the coming years. 

We have seen this pattern before with the rise of online trading in the 1990s when a raft of online trading companies boldly predicted that they would threaten the existing retail brokerage model.  E*Trade’s CEO told the industry that online brokers would destroy Merrill Lynch, which he described as a dinosaur.  Today, E*Trade is a subsidiary of Morgan Stanley, virtually none of the online trading firms of the 90s survived, and every bank or broker today offers online trading.

“DeFi is not going to replace JP Morgan, but JP Morgan is going to use a lot more blockchain.”

Tokenization Needs Infrastructure. Hardware Security Has Spent Decades Building It

As mainstream finance arrives onchain, it brings assets that do not look like what we think of as “crypto.” That is what changed my view of Ledger. When the company first called, I saw a crypto company with a loyal self-custody audience. Most tokenized assets so far have been crypto, but now money market funds, stocks, bonds and deposits are also going onchain, and neither fits the old picture. Within the next one to two years, we will stop saying crypto and start saying digital assets.

Anything onchain needs far stronger security infrastructure, and hardware security has been providing that kind of infrastructure for a long time. France has deep roots in that field, going back to the chip in your bank card. Ledger carries that tradition forward, and it is a different discipline from software. As I told Ronit Ghose: “You can be a software engineer, you can be a hardware engineer. They’re not interchangeable.”

Tokenization Removes the Safety Net That Slowness Provided

Traditional payments get criticized for being slow. A wire can take days, and that is a real flaw for customers. But that latency offers security: payments are reversible, and someone has time to spot a mistake. In 2016, that delay saved Bangladesh Bank over $800 million.

The Drift hack is the counterexample, where the assets were not entries on a third party database, but cryptographic keys. Control the key and you control the asset and that’s exactly what North Korea’s Lazarus Group did with Drift, in 12 minutes.

“When something’s tokenized, it becomes exploitable at machine speed.”

AI adds additional dynamics: autonomous agents will increasingly transact on our behalf and AI is empowering bad actors to attack our cyber defenses with unprecedented speed and scale.

Our CEO Pascal Gauthier made the case for hardware over software in his essay Revenge of the Atoms. We have relied for decades on software defenses, which are walls of code. But every day we see those walls are crumbling under an AI powered onslaught. The tokenized/AI era we are fast entering needs an alternative – walls of silicon: protect the keys and control the underlying asset, and encode the governance that controls transaction approvals in hardware.

The bank clerk in 2016 that inadvertently stopped the Bangladesh Bank hack was a checkpoint in the payments system. Not one designed for security, but in this case, one that became the final security bulwark. Transacting in an era of growing autonomous agents will need checks that are resistant to exploits. That is the idea behind Ledger Agent Intent: agents propose, humans approve, and hardware enforces. Our 2026 AI security roadmap and Ian Rogers’s Security in the AI Agent Era lay out the approach in more detail.

Ledger Enterprise already works this way for institutions. Governance rules, approval workflows, and transaction policies are enforced in hardware security modules and controlled by the customer’s own administrators, who approve with Personal Security Devices (PSDs). Ledger’s Agent Stack brings that layering to agents, and Ledger Agent Intent serves a different user. An individual has no governance team and no HSM, so the control is simpler and personal: you review each proposal your agent makes and confirm it on your own Ledger signer.

All of it rests on keys.

“Literally everything we do that matters uses cryptography.”

Quantum Is a Deadline Without a Date

Gaurav found the natural segue to the discussion around cryptography – post quantum cryptography (PQC), which I think naturally follows from the dependence on keys. A machine that breaks today’s cryptography breaks everything built on it, and nobody can say when such a machine will exist.

“Quantum is one of the least appreciated challenges we face, because it’s like Y2K but without a definitive date.”

Every chip in every passport, bank card, and industrial control system will need an upgrade, and we have not really begun. Our CTO Charles Guillemet explained how that impacts Bitcoin in a September post: Bitcoin does not have a quantum computer problem today, but it has a migration problem, and migrations take years. But PQC is a much broader challenge than “crypto.” Cryptography underpins almost anything of consequence we do today. Communications, financial transactions, industrial control systems all use cryptography and will need to be upgraded for a PQC era. That means new cryptographic algorithms that will produce larger cryptographic signature sizes than the typical 64-256 bytes of today.  Signature sizes of the algorithms thus far approved by NIST are typically in the 2-4KB range with some producing signatures as high as 50KBs. These will require all those Secure Element chips in our credit cards and passports to be replaced and upgraded.  A huge task that nobody is yet undertaking.  

Inevitable Does Not Mean Universal

As the conversation moved on, Ronit asked what an ordinary senior banker, not a tokenization specialist, says to me about all this. It was a fair challenge, and my answer was a warning about absolutes.

“We tend to think too much in absolutism.”

My analogy is US television, where broadcast competed against cable, then both had to compete against streaming. But all three still exist. In ten years, a lot of payments will run on blockchain, perhaps the majority, but today’s rails like ACH, FedNow and SWIFT will still exist. To this day, I still keep a checkbook.

Blockchain speed will take hold first in enterprise payments and Wall Street trading, which have always been early adopters. But there will be resistance from the innovator’s dilemma.  Many major businesses inside banks will be disrupted by blockchain the same way American Express lost traveler’s check revenue, the same way the spread of globally connected ATMs and AT&T lost long-distance revenue from the rise of VoIP. Those businesses will not go quietly into the night, but blockchain’s efficiency will make its adoption inevitable over time.  

Security Infrastructure Is Where Ledger Is Headed

The slice that does move onchain needs security built for it. When Ronit asked where Ledger goes in three to five years, I described a move from securing crypto for self-custody users toward security infrastructure for tokenized, AI-driven systems. Identity is the use case I reached for: an identity held in a secure hardware device, rather than in a cloud password manager that has to be trusted never to be breached.

In recent months, Ledger has announced two products that point the same way. At Ledger Op3n on October 15, the company introduced Ledger Atom™, a security and privacy device that runs everyday apps with a secure chip, a secure screen, and Ledger OS, and Ledger Agent Intent, a single interface where agent approval requests land. 

Hardware security is tested the hard way at Ledger Donjon, our white-hat hacker lab, where researchers attack devices with electron microscopes and lasers (donjon.ledger.com). The name is French, it translates as ‘keep’, which is the most secure part of a castle, a great metaphor for a Secure Element, the component at the heart of Ledger signers, and a reminder that when money moves at machine speed, security has to live in hardware.

---

John Andrews, Ledger CFO