When Google Quantum AI reported in July that it had taught a quantum computer to “learn from its errors,” the obvious story was better error correction. Quantum computers are extremely sensitive machines. Small changes in their environment can push them out of tune and introduce errors, which can force researchers to stop a calculation and recalibrate the hardware.
Google’s AI system learned to make some of those adjustments while the machine was still running. In tests, it made the system 3.5 times more stable when researchers deliberately knocked it out of tune and reduced errors by another 20% after human experts had already calibrated it.
Useful quantum calculations may eventually need to run continuously for days or even months. Keeping the machine stable without repeatedly stopping the calculation therefore matters. Google likens it to tuning an instrument while the music plays.
Sustained stability is also what a future quantum computer would need to run Shor’s algorithm long enough to attack widely used public-key cryptography.
The more consequential development is what AI is doing to the pace of quantum research itself. Error correction, circuit design, calibration, and cryptanalysis used to improve on partly separate timelines. AI is starting to compress several of those research loops at once. Quantum research is now moving at software development speed. Cryptographic migration moves on a slower clock, set by deployed hardware, vendors, governance and users.
Large-scale cryptographic attacks remain beyond current hardware, yet researchers keep reducing the amount of quantum computing power that an attacker would need to break today’s vulnerable cryptography.
The Target Keeps Getting Smaller
Peter Shor showed decades ago that a sufficiently capable quantum computer could break RSA and elliptic-curve cryptography, two widely used forms of public-key cryptography that protect digital communications. The mathematics has been known for decades. The uncertainty has always been when the hardware becomes capable enough to run the attack at a useful scale.
In March, Google Quantum AI published new estimates for attacking widely used elliptic-curve cryptography. The researchers calculated that Shor’s algorithm could run with fewer than 500,000 physical qubits under the hardware assumptions they tested. A previous leading estimate required roughly 20 times more. No equivalent 20-fold hardware leap had occurred. Better circuit design reduced what the hardware eventually needs to do.
The same dynamic became visible through ECDSA.fail, an open effort where researchers and AI agents searched for ways to make the same quantum attack require less computing power. The network matched Google’s benchmark within hours and surpassed it within days. By the end of June, the best reported circuit was 47.5% more efficient on the benchmark.
Circuit efficiency and calendar time move differently. Every efficiency gain still reduces what future hardware has to do, and AI can keep pushing that requirement down before institutions have finished reacting to the previous estimate.
People naturally ask when the code-breaking machine arrives. For security planning, I think the more useful question is how quickly our assumptions can become stale.
Defense Runs on a Different Clock
NIST finalized its first post-quantum standards in 2024. Cloudflare says more than 65% of human traffic on its network already uses post-quantum encryption. Apple deployed a post-quantum cryptographic protocol called PQ3 for iMessage, and Signal has continued adding post-quantum protections to its protocol.
Anthropic’s July work shows how AI can strengthen the defensive side as well. Claude found a significant weakness in HAWK, a post-quantum signature candidate that had already survived two rounds of human review. Because HAWK was still a candidate, researchers found the weakness before deployment, and NIST’s finalized standards were unaffected.
Once researchers agree on better cryptography, organizations still have to find where old systems remain, coordinate vendors, update hardware and software, and move customers who control their own keys. Those dependencies can take years. Post-quantum migration is an infrastructure program with a cryptographic deadline that nobody controls.
Blockchains Make the Gap Harder to Hide
Most leading blockchains rely on a pair of cryptographic keys to establish ownership. The public key can be visible to the network. The private key is the secret proof that allows the owner to move the assets. Bitcoin uses this key pair to authorize transactions.
A sufficiently capable quantum computer running Shor’s algorithm could work backward from an exposed public key and calculate the private one. At that point, an attacker could produce a valid transaction and move the funds as though they were the legitimate owner.
To the blockchain, that forged transaction would look legitimate. There is no help desk that can reverse the transfer after the network accepts it.
Web services have more control over migration. Apps like Signal can simply ship a protocol update. Cloud providers can alter key exchange behind the scenes, and most users may never know anything changed.
A decentralized network has to coordinate wallets, exchanges, custodians, validators, developers, hardware manufacturers, and token holders. Some keys are already lost. Some addresses may remain untouched for years. Post-quantum signatures also require substantially more data than the digital signatures blockchains use today. Every transaction therefore takes up more network and storage capacity, which can reduce the number of transactions a blockchain can process efficiently.
Migration also requires existing balances to move to quantum-secure addresses. Hardware wallets must support the new cryptography; exchanges must recognize the new addresses; custodians must update their systems and policies. Dormant or lost funds may never migrate, which leaves the network with governance decisions that cryptography cannot answer.
I think the industry still underestimates how long that coordination can take. The cryptography can be ready while the network around it still needs years to migrate, because decentralization removes the authority that could simply order everyone onto a new standard.
Start Measuring Migration Time Now
Blockchain security teams should be able to answer a basic operational question today: if the migration deadline suddenly moved forward by three years, how long would it take to get users, wallets, exchanges, and critical protocol infrastructure onto post-quantum signatures?
For many networks, the answer is still unclear. Another debate over whether Q-Day lands in 2030, 2035, or later will not resolve that uncertainty.
To prepare, identify where vulnerable cryptography is still in use. Benchmark post-quantum alternatives under real network conditions. Build wallet and custody support before it becomes an emergency feature. Start the governance work while disagreements can still take months instead of hours.
AI is already shortening the research cycle on the quantum side. Post-quantum algorithms protect assets only after wallets, exchanges, and protocols can actually use them. Security teams should build that migration path before a quantum breakthrough turns planning into incident response.
