An AI-assisted quantum error-correction breakthrough could sharply reduce the qubit threshold needed to threaten today’s internet encryption.
A Caltech spinoff called Oratomic, working alongside Google Quantum AI, used a large language model to discover an error-correction algorithm that reduces the qubits needed to break standard internet encryption from millions to as few as 10,000.
The finding, published March 30, compresses the timeline for quantum-capable attacks on RSA and elliptic curve encryption by an estimated decade – and has triggered an accelerated response from major internet infrastructure providers.
Why It Matters
The number that matters is the reduction factor: roughly 100 times fewer physical qubits per logical qubit, achieved by applying AI to discover quantum low-density parity-check error-correction codes. Previous estimates required millions of qubits to crack P-256 encryption. The new calculation puts the threshold at 9,988 qubits running for approximately 1,000 days, or 26,000 qubits completing the task in a single day.
Oratomic CEO Dolev Bluvstein said the role of AI was unambiguous: “There is no question that we used AI to accelerate this development. No question at all.” He added that the world is “currently not prepared” for what the timeline now implies. Cloudflare’s Bas Westerbaan called the findings “a real shock” and announced the company is moving its post-quantum migration deadline forward to 2029.
For the AI Research community, this story represents something distinct from model benchmarks and product launches: AI functioning as a scientific accelerant in a field it wasn’t built to solve.
What’s Next
Neither paper has completed peer review, and Princeton’s Jeff Thompson flagged that the results rely on “aggressive assumptions about the speed of operations.” A follow-up paper detailing the AI methodology is promised but not yet published. Until independent validation arrives, the findings should be treated as directionally alarming rather than operationally definitive.
Governments and enterprises with long data-protection horizons – finance, defense, healthcare – can’t afford to wait for peer review to begin migration planning. NIST’s post-quantum cryptography standards finalized in 2024 are already available; the bottleneck is implementation speed, not the existence of alternatives.
Google’s separate paper showing Bitcoin’s secp256k1 encryption could fall to fewer than 500,000 physical qubits adds a second independent data point. Two teams arriving at structurally similar conclusions through different methods is harder to dismiss than one. The quantum threat window has likely closed from decades to years.
Sources: TIME · Nature · Quanta Magazine · Science News
About the Author
