By Dr. Leandro Aolita, Chief Researcher, Quantum Research Center, Technology Innovation Institute
In August 2026, the US Treasury launched a Quantum-Readiness Task Force, following an executive order signed in June directing federal agencies to strengthen cryptographic protections. The task force brings together government agencies, financial institutions and technology vendors to move the US financial sector onto post-quantum cryptography. It is one of a series of measures, in Washington and elsewhere, that now treat quantum risk as a live planning problem rather than a distant hypothetical.
The urgency is reasonable. Cryptographic transitions on this scale take years to plan and execute, and a government that waits for certainty about the threat will have waited too long. Less examined is the question beneath the policy: What has to happen, technologically, before a quantum computer capable of that kind of threat exists at all?
Machines with more than a thousand physical qubits are no longer remarkable. The largest number of error-corrected logical qubits ever demonstrated is 96, reported by Nature in January, running on 448 physical atoms. But qubit counts alone do not determine security relevance. What matters is how many logical qubits a system can sustain, how many operations they can perform, and whether the error rates are low enough.
Individual quantum bits, or qubits, are extraordinarily fragile. Stray heat, electrical noise or a slightly imperfect control pulse can corrupt the information they hold. Errors do not simply accumulate as a computation proceeds; they compound across operations, so that a long calculation on unprotected qubits can become unreliable. A machine that makes too many mistakes cannot be trusted with anything that matters, whatever its qubit count.
Quantum error correction addresses decoherence by combining many noisy physical qubits into a single, sturdier logical qubit. Repeated rounds of measurement, known as syndrome measurements, check for signs of error without disturbing the information being protected, while a separate step, decoding, works out from those measurements what most likely went wrong.
The obstacle is overhead. Larger, more reliable logical qubits generally mean more physical qubits, more syndrome measurements and more decoding, all of it running fast enough to keep pace with a live computation.
Advances in quantum error correction will determine when a cryptographically relevant quantum computer becomes a real threat. Error correction cannot be achieved simply by adding more qubits; those qubits only help if they can be controlled and corrected reliably.
Two recent papers from our group at the Quantum Research Center start from the same observation. Quantum computers already generate a great deal of information about their own errors, and that information could be used more.
The first paper applies machine learning to the syndrome measurements a system is already collecting to identify which runs of a computation are heading for failure. The most accurate existing methods for this rely on detailed, code-specific calculations from the decoder, which are expensive to run. Our approach instead trains a classifier directly on syndrome data alone, teaching it to distinguish low-noise runs from high-noise ones without that overhead. In simulation across two families of error-correcting code it matched existing filtering methods; on experimental data from a neutral-atom processor, it outperformed them when combined with the original method.
The second challenges a more basic assumption: that syndrome measurements should occur at a fixed interval, like a metronome. Measure too rarely and idling errors accumulate; measure too often and the measurements themselves introduce faults. We show, analytically and in simulation, that the optimal interval depends on the size of the code, and that getting the timing right, or adapting it as noise conditions change, reduces logical error rates by a margin that grows exponentially as codes get larger. Applied to noise parameters published by Google from its own error-correction experiments, the model indicates that timing optimization alone could cut logical error rates per unit time by up to 40 percent, with no change to the hardware.
Neither result, on its own, establishes when a cryptographically relevant quantum computer will become available. But both bear on how the policy and scientific conversations should meet. Governments are right to prepare before the threat materializes. Being able to distinguish genuine progress in quantum reliability from a qubit-count headline is what allows policymakers, security teams and boards to calibrate that preparation: to take the risk seriously without reacting to every hardware announcement.
The next phase of quantum computing will not be settled by qubit count alone, but by how effectively machines can detect and correct errors. Progress in error correction will help determine how soon quantum computers can threaten today’s cryptography—and, with it, the timeline for everything else
