A remote server tells you it just rolled a die and got a six. How do you know it actually rolled? Maybe its "randomness" comes from hard-coded pseudo-random code, or the seed was compromised long ago. Much of the modern world runs on trust in randomness: lotteries, blockchain consensus, cryptographic key generation, audit sampling. If randomness becomes predictable, the foundation crumbles.
On September 30, a paper appeared on arXiv by Siddhartha Jain, Saachi Mutreja, and Bhaskar Roberts. It solves a long-acknowledged open problem in certified randomness — and fills exactly the last gap: making quantum-generated randomness both verifiable and as "full" as theoretically possible.
What "certified" means
The classic dilemma: how do you prove to a stranger across a network that your randomness isn't faked? Ordinary approaches all rely on trust — in the hardware vendor, the data center, luck. Certified randomness flips the logic: don't trust the device, verify the math.
A classical supercomputer acts as examiner, giving the quantum computer a problem it excels at — random circuit sampling — which the classical machine can also solve, only absurdly slowly. When the quantum device submits its answer and it passes verification, the key insight kicks in: since the classical examiner could not possibly have produced those bits that fast, the bits must have been freshly generated by the quantum device, on the spot. A cheater either can't compute the answer or must actually own the quantum machine. The randomness is mathematically stamped.
This line of work began with Scott Aaronson's 2018 proposal; in 2023 he and Shih-Han Hung turned it into a rigorous protocol. In March 2025 it left the page for the first time: JPMorgan, Quantinuum, Argonne and Oak Ridge National Laboratories, and UT Austin ran the protocol on the 56-qubit H2-1 trapped-ion quantum computer, with Argonne's Polaris supercomputer as examiner, published in *Nature*. It is frequently cited as the strongest candidate for quantum computing's "first real application."
But the glass was only half full
Two gaps remained. The first is rate: certified randomness quality is measured by the ratio of min-entropy to bit length — the closer to one, the purer. Existing protocols typically yielded weak randomness, with the ratio approaching zero, like wine diluted with ninety percent water; post-processing was required. The second is worse: some protocols require the verifier to first supply trusted seed randomness. To prove your die is real, you need a die everyone already trusts. Trust wasn't eliminated, merely relocated.
The new paper closes both gaps at once. In the quantum random oracle model — a cryptographic framework — the protocol's security is unconditional within the model, produces randomness at rates approaching the theoretical maximum, and requires no trusted randomness from the verifier at any point. It improves on prior results of Coladangelo et al. and directly answers the open question left by Aaronson and Hung. The paper also delivers a new primitive — a conditional min-entropy proof: even if an adversary sees all previously published outputs, the next batch of random bits remains unpredictable. For randomness beacons that continuously "broadcast" randomness, this is essential.
The simplifications, stated honestly
Two details were compressed. First, the "quantum random oracle model" is an idealized framework treating hash functions as perfect random functions — an excellent approximation in practice, but strictly speaking security is unconditional within the model, not literally unconditional. Second, the jump in rate from near-zero to near-one is theoretical; engineering a beacon spewing millions of bits per second is still distant — in the 2025 experiment, the quantum machine took several seconds per challenge. The paper doesn't compute those numbers, and nobody has yet.
A source of randomness requiring trust in no one means lotteries can be audited publicly, consensus can be reached among mutually distrustful machines, and key generation can be verified end to end. The theoretical blueprint has drawn its final stroke. Which data center first turns it into a business producing a billion certified bits a day — that race hasn't started yet.