In Brazil, a medical record must be kept for 20 years. Will the encryption protecting it last that long?

When did the breach happen?

Imagine your company has a secret.

It is protected today. The controls work, the technology in use is considered secure and the last audit found nothing of note.

Now imagine someone copies that secret. They can’t read it, can’t use it, can’t even make sense of the contents. So they simply keep it.

For ten years.

Until a technology that didn’t exist on the day of the theft makes it possible to read everything.

In that scenario, when did the breach happen? When the information was copied? When it was finally decrypted? Or years earlier, when someone decided that the protection available at the time would be enough for the entire life of that secret?

This is not only a debate about post-quantum cryptography. It is about an uncomfortable feature of technology: the future can change the risk of decisions we make in the present. And companies are particularly bad at thinking this way.

The attack that starts with a copy

The scenario above has a name, and it already appears in official documents. NIST, the US standards institute, calls it "harvest now, decrypt later". In a November 2024 document, the institute writes that encrypted data remains at risk because adversaries can collect it today and store it until they have the means to decrypt it, and it cites government secrets and medical records as examples.

The White House report on post-quantum cryptography, from July 2024, describes the same problem in a way I find even clearer for non-technical readers: encryption does not stop anyone from making an exact copy of the encrypted data. In Europe, the migration roadmap published by the member states in June 2025 lists government data, sensitive personal data and trade secrets among the targets.

Notice what this does to the idea of a leak. At most companies, the incident begins when someone reads what they shouldn’t. Here, the incident begins with a copy that, on the day, causes no harm at all.

The half-life of a secret

In 2015, the Canadian researcher Michele Mosca wrote a paper that became a reference on the subject. One of his variables is what he called security shelf life: how long a piece of information must stay protected. His examples range from 10 to 20 years and reach 100 years, for things like health information, trade secrets and national security data.

Brazilian law offers some concrete numbers for that shelf life. Law 13.787 of 2018 requires a patient’s medical record to be kept for at least 20 years from the last entry. The Access to Information Law allows a public document to be classified as top secret for 25 years, a term that can be renewed once. And then there are the secrets no law accounts for: the board minutes on an acquisition, the pricing strategy for a new product line, research that has not yet become a patent application, the negotiation of a long-term contract.

Some secrets lose their value tomorrow. Others remain sensitive for decades. And almost every data classification scheme I know treats both the same way: "confidential".

When the future arrives

No one knows the date on which a quantum computer will be able to break the encryption used today, and it is only honest to say so up front. What we have are estimates, and they keep getting shorter.

The Global Risk Institute runs a periodic survey of experts in the field. In the seventh edition, for 2025, released in March 2026, the 26 experts surveyed considered it "quite possible" (28% to 49%) that a quantum computer capable of breaking current encryption will exist within ten years, and "likely" (51% to 70%) within fifteen.

On the technical side, the math has also become cheaper. In 2019, Craig Gidney of Google Quantum AI estimated with Martin Ekerå that breaking a 2,048-bit RSA key would require 20 million noisy qubits running for about eight hours. In May 2025, Gidney published a new estimate: fewer than one million qubits, in less than a week. Today’s computers are still far from that. But the number of qubits required fell roughly twentyfold in six years.

If a secret must last 20 years and the experts consider it likely that the threat arrives in 15, the math no longer works in the secret’s favour.

The detail that should trouble anyone in charge of data

The NCSC, the UK’s cybersecurity agency, published its timeline for migrating to post-quantum cryptography in March 2025. Among the technical milestones is a request that sounds simple: that organisations record the expected lifetime of the data they hold.

I have never seen that field in a data classification policy. We have "public", "internal", "confidential", "restricted". We don’t have "must stay secret until when". And without that answer, there is no way to know whether the protection chosen today will last long enough.

What changes when a secret has an expiry date

1. Classify by half-life, not just by sensitivity. For every type of sensitive data, one extra question: how long does this need to stay secret? Five years, twenty, fifty. The answer changes the priority of everything.

2. Protect what lasts longest first. Long-lived data travelling across the network is the natural target for anyone collecting today to decrypt tomorrow. That is where it makes sense to start using post-quantum algorithms, which NIST standardised in 2024 and which are already available in browsers and libraries.

3. Keep less. The best-protected secret is the one that isn’t stored. Every piece of sensitive data kept longer than necessary is one more secret waiting for the right technology.

The question for the next set of minutes

The next time a strategic document is produced, whether board minutes, an acquisition memo or a research project, ask one question before filing it: how long does this need to stay secret, and does the protection we are using last that long?

Perhaps the most important breach of the next decade is already under way. Not in the copy, which no one will notice. In the decision to believe that today’s protection is enough forever.