Published October 2026.

On 27 October 2026, NIST published NISTIR 8547, Transition to Post-Quantum Cryptography Standards. It is a planning document, and that is exactly why it matters: it turns "someday" into dates.

It also carries a lesson that reaches well beyond cryptography. When a threat is clearly on its way, waiting is the risky choice. AI security is facing the same test right now.

What NIST Has Published

NIST released its post-quantum cryptography standards in August 2024, including FIPS 203, which covers key establishment. Those standards answered the question of what to migrate to. The new report answers how and when: it sets out NIST's official approach and timelines for moving off quantum-vulnerable algorithms.

The direction is clear. Algorithms like RSA and ECDSA will be phased out of federal standards. Some will be disallowed after a set date. Others will remain allowed only at the user's own risk, which is a polite way of saying the responsibility moves to you.

Why the Clock Started Years Ago

The urgency comes from a simple attacker strategy: harvest now, decrypt later. An adversary does not need a working quantum computer today. They only need to collect encrypted traffic and stored data now, then wait until quantum computers are capable enough to break it. Any data that has to stay confidential for years is already exposed to that plan.

That is why the deadlines matter. National Security Memorandum 10 set 2035 as the target for finishing the migration across federal systems. Executive Order 14412 pulled that forward for federal high-value assets and high-impact systems: 2030 for key establishment and 2031 for digital signatures.

The Practical Reality of Migration

Two points in the report deserve attention from anyone planning this work.

First, hybrid approaches are accepted. You can combine a classical algorithm with a post-quantum one, like putting two padlocks on a door: an attacker has to break both. The trade-off is cost, because running two schemes takes more compute.

Second, the migration covers the whole cryptographic supply chain, not just the systems you build yourself. Libraries, vendors, devices, protocols and partners all have to move with you. NIST's National Cybersecurity Center of Excellence (NCCoE) runs a project with practical demonstrations that show what this looks like in real environments.

History is not encouraging. Past cryptographic migrations have taken 10 to 20 years. Measured against that, 2030 is very close, and technology alone will not close the gap. People have to understand the risk, find where vulnerable cryptography lives, and do the work.

AI Security Is Running the Same Race

While organisations plan for a quantum future, AI is reshaping their attack surface today. AI adoption is growing fast, and so is the list of security problems that come with it:

Prompt injection, where crafted input or poisoned content overrides a model's instructions.
Data leakage, where a model reveals its system prompt, internal data or information it was never meant to share.
Manipulated or abused models, from poisoned data to assistants turned against the people they serve.
More attack surface with every AI feature, because each new assistant, plugin or agent is another way in.

The parallel with post-quantum cryptography is direct. In both cases the threat is real, the fixes take time to roll out, and the organisations that wait for a crisis will find themselves years behind it. There is even a version of harvest now, decrypt later: a system prompt or customer record leaked through an AI feature today stays leaked.

Preparation Is a People Problem

New standards and better tools are necessary, but they are not enough. Cryptographic migrations stall when teams do not know where the risk sits. AI security fails the same way when developers ship features without understanding how those features can be attacked.

That makes training and AI security awareness essential, not optional. Reading about prompt injection is a start. Real confidence comes from practice: attacking an LLM system yourself, watching a defense fail, and then building one that holds. Teams that have done this hands-on spot risky designs early, ask better questions in review, and respond faster when something goes wrong.

Start Practising Now

QbitDojo gives you hands-on labs where you attack live AI targets and capture the flag. Begin with The First Override, a free beginner lab built around the classic "ignore previous instructions" attack. Move on to Data Smuggler, where you hide instructions inside a data field the model is meant to process, and later to Poisoned Context, where you plant instructions in retrieved data the AI trusts as authoritative.

Then switch sides. In Defense Mode you write the hardened system prompt that should stop the attack, and an AI evaluator scores whether it blocks the attack while keeping the assistant useful.

The quantum deadlines give organisations a few years. AI attackers are not waiting at all.

Start your first lab

NIST's Post-Quantum Roadmap and the Case for Preparing AI Security Now

Comments

Comments