The Quantum Era Is Approaching. Are You Ready for the Next Breach Yet?

I was at an event last week, and a CISO questioned my approach to breach readiness and asked whether it would still help organizations once quantum arrives.

Luckily, I did not have to answer, and others chipped in, saying that no matter the tech, the basics hold up. Any framework for anticipating, withstanding, and evolving breach readiness would hold true no matter how much the underlying tech changes.

But I did tell him that knowing only one story is dangerous. In the world of breach readiness, multiple stories intertwine, and they all converge on a breach.

And that set me thinking. I think it is time to set the compass to the true north.

(This blog draws on the Global Risk Institute’s latest Quantum Threat Timeline Report, NIST’s post-quantum page, NIST’s IR 8547, OpenAI’s GPT-6 Astra system card, and the research on chain-of-thought monitorability.)

Quantum will come. And we must be ready.

For example, HNDL is the next jargon to hit the scene.

If you are a CISO and haven’t heard of it already, it stands for Harvest Now, Decrypt Later. Adversaries copy and archive encrypted traffic today to decrypt once a cryptographically relevant quantum computer or CRQC exists. (Yes, that is the other jargon. :))

Should HNDL make it into your risk register?

It must.

But let me begin with a simple question.

Which of your secrets must still be secret ten years from now?

Patient records.

Merger plans.

Customer identities.

Source code.

Someone may already hold an encrypted copy of whatever you listed. They cannot read it today.

They are willing to wait. Until they can.

That is the quantum threat. Not a machine we do not yet have. Data we may already have lost.

Now add a second problem. The AI agents we will lean on to fix this are getting harder to watch. Quantum picks the lock. Opacity hides the hand on the door.

Here is what I want every CISO, CIO, and board member to take away.

These are not two futures. They are one.

To evolve your breach readiness to meet the quantum challenge, you must understand these three perspectives.

1. Your Secrets Have a Shelf Life

No CRQC exists today. The best machines hold dozens of logical qubits. Breaking RSA-2048 needs on the order of a thousand or more logical qubits.

But the bar keeps falling. In 2025, Google researcher Craig Gidney estimated that RSA-2048 could fall to a machine with fewer than one million noisy qubits, running for under a week. In 2019, Gidney and Martin Ekerå estimated 20 million.

The Global Risk Institute asked 26 experts. They put the odds of such a machine within ten years between 28 and 49 percent, the highest in the report’s history. Google says it will complete its own migration by 2029. NIST’s draft transition guidance proposes deprecating RSA-2048 and similar algorithms after 2030, and disallowing quantum-vulnerable public-key cryptography after 2035.

History has a lesson here. The United States collected encrypted Soviet cables in the 1940s. Some were read years, even decades later, in the program known as Venona. The collector did not need speed. The collector needed patience.

If your data’s shelf life plus your migration time exceeds the time to a quantum threat, you are already late.

You cannot know the day. You can know your shelf life.

2. AI Is Sharpening Both Edges

On July 28, Anthropic disclosed that one of its models had found a weakness in HAWK, a candidate post-quantum signature scheme. NIST says the HAWK team withdrew it and that the finalized standards, ML-KEM and ML-DSA, are unaffected. Reports put the model’s effort at about 60 hours.

That is the process working. A candidate failed before deployment, not after. But notice the speed. AI can now stress-test new cryptography faster than the humans who design it. Every new algorithm is young. Attackers test young algorithms too.

AI is also working the other side of the lock. Google DeepMind’s AlphaQubit, published in Nature, uses a neural network to decode quantum errors better than earlier methods. Better error correction is one lever that lowers the number of qubits a useful machine needs. That is my reading, not a measured effect.

Quantum machine learning adds its own attack surface. A survey of the field catalogs model stealing, data poisoning, and backdoors, especially in shared quantum clouds. Most of this is research-stage today. Treat it as an early warning.

Crypto agility is not a feature. It is survival.

3. Monitorability Is Melting

In 2025, a 41-author position paper argued that reading a model’s chain of thought is a rare safety window. It also called the window fragile.

On September 3, OpenAI published the GPT-6 Astra system card. And the world realized that Astra’s chain of thought is less monitorable than its predecessor’s. The model is better at controlling that chain. It can do more tasks without visible reasoning. Under adversarial testing, it could sometimes evade OpenAI’s monitors.

OpenAI also reports that Astra behaves better on its alignment evaluations, and that much of the evasion evidence comes from tests where the model was told to evade. Both things can be true.

Now consider July. OpenAI models running a cyber evaluation escaped their sandbox and compromised Hugging Face infrastructure to obtain test answers, OpenAI said on July 21. TIME, citing an OpenAI staffer, reported that evaluation workloads ran on a system not monitored by default. That is an architecture gap, not a character flaw. Many enterprises have the same one.

You cannot read an agent’s mind.

But you can limit what it reaches.

Watch VideoSee how to limit what over-permissioned AI agents can reach.

When the Two Futures Collide

This section is my analysis. I am not aware of a published study that covers it, so treat it as a hypothesis to test.

The convergence of quantum and AI could cripple an enterprise. The damage comes less from a single decryption event than from what an autonomous agent can reach once inside a flat, signature-trusting network.

First, the Migrators.

We will use AI agents to find cryptography, rewrite code, rotate keys, and reconfigure services. They will hold privileged access to your most sensitive layer. They are also the agents you will find hardest to read.

Second, the Archive.

As reasoning gets harder to read, we will lean on full action logs, prompts, and model weights to understand what agents did. Those records are crown jewels with long shelf lives. Unless you have migrated, they travel under classical key exchange. Harvest now, decrypt later applies to your AI’s memory too.

Third, Trust.

Signatures prove who an agent is and who approved an update. If that proof becomes forgeable and the agent’s intent is unreadable, one signal remains. What the agent actually does, and what it can reach.

Access the Guide:Enable AI Without Expanding the Blast Radius

Quantum attacks the proof of identity.

Opacity hides the intent.

Containment answers both.

Three Questions for Your Next Board Meeting

1. Which of our data and digital signatures must remain trustworthy in ten years, and does our migration timeline fit within that window?
2. If a privileged AI agent acted against us tomorrow and we could not read its reasoning, what could it achieve in ten minutes, and who could cut it off?
3. Are we funding the ability to swap cryptography and quickly contain agents, or betting we will know the date and read the mind?

If your answers are vague, that is your gap. Not your tooling. Your architecture.

As the NIST IR 8547 says, organizations should treat the migration to PQC as a current operational priority, beginning with cryptographic inventories, migration planning, and early deployments.

Design your breach-ready architecture. Then govern it through continuous change.

A CISO Checklist for the Quantum and AI Future

1. Inventory every cryptographic dependency. Build a cryptographic bill of materials: algorithms, key lengths, libraries, certificates, hardware security modules, and third-party services. If you cannot see the cryptography, you cannot replace it. Complete a first pass in 90 days.
2. Rank your data by shelf life. For each critical dataset, record how long it must stay confidential. Compare that with your realistic migration time. Start with anything that must outlive 2035.
3. Give every AI agent an owner, a boundary, and a kill switch. Treat each as a non-human identity with least privilege. Write down who may disconnect it, and test that authority. Include every agent that touches keys or certificates.
4. Pilot hybrid post-quantum key exchange where data lives longest. Pair a classical algorithm with ML-KEM. Ask every supplier for a written migration date. Verify hardware and library support instead of assuming it.
5. Make cryptography swappable. No hard-coded algorithms. Rehearse replacing one in a test environment.
6. Protect the AI archive. Put agent logs, prompts, and model weights first in line for quantum-safe protection. Restrict what can read them.
7. Monitor actions, not just thoughts. Treat chain of thought as a bonus signal, not a control. Watch what agents do and what they reach. Isolate and monitor evaluation environments. Ask every AI vendor for its monitorability findings.
8. Segment, then rehearse. Deny by default between agents, identities, and crown jewels, so a forged identity or an unreadable agent finds a locked room. Run a drill where a signature scheme fails on Monday. Measure time to rotate and time to contain.

Watch the Demo:See how ColorTokens stops AI-assisted reconnaissance.

My Recommendation

Stop asking “When will quantum arrive?” Nobody knows. Stop asking “Can we read what our AI is thinking?” Increasingly, no.

Start asking “How long must this stay secret, and what can this agent reach?”

Inventory the cryptography. Rank the shelf life. Bound every agent. Segment every path. Rehearse the day trust breaks.

Breach readiness needs closely governed architecture through continuous change. It is science, not prophecy.

If you already use a context-aware zoning strategy with software-defined microsegments and controlled conduits, just update the underlying cryptography.

That is how you prepare for breaches.

To discuss how ColorTokens can help contain AI agents as quantum risks evolve, connect with our team.