A cryptography paper accepted to Crypto 2026 pulls off something narrow, and is refreshingly upfront about it: it deletes a clock.
Not the wall-mounted kind. What goes is a randomness beacon — the always-on service that hands every participant in a protocol a fresh public random value at regular intervals. Earlier constructions in this line of work simply assumed such a service was there. This one argues you can do without it.
What the paper actually removes
Damiano Abram, Marshall Ball, Juan Garay and Aggelos Kiayias are credited as authors of “Permissionless consensus from a common random string,” according to the University of Edinburgh research record. Crypto 2026 runs Aug. 17 through Aug. 20 in Santa Barbara.
The question they take on is old and stubborn. How do parties reach Byzantine agreement when nobody knows in advance who is taking part, beyond an upper bound on the number of them, and there is no public-key infrastructure to lean on? Bitcoin helped motivate this research line. But nothing in the public record casts the new work as a Bitcoin upgrade or a change to any live network, and anyone telling you otherwise this week deserves suspicion. The 2024 version needed a heartbeat
Back in 2024, Ball and collaborators put together a construction that paired proofs of work with fine-grained complexity assumptions. Its model also handed every participant a beacon that delivered a fresh public random value at regular intervals. That is the recurring service — the heartbeat the whole design was quietly synced to.
In its place, the new paper puts d-wise independent distributed samplers. Plainly stated, the samplers are designed to keep multiple executions secure simultaneously, with no beacon obliged to keep pumping out new public randomness on a schedule. Setup doesn’t disappear, it changes shape
Here is the piece that tends to fall away as a result like this travels. Dropping the beacon is not the same as dropping assumptions.
A common random string is still in play for participants. Per the abstract, it need not be structured, nor sampled precisely at the moment the protocol starts — a meaningful loosening rather than an elimination. The sampler constructions lean on the decisional Diffie-Hellman and learning with errors assumptions. Those samplers are then combined with proofs of work grounded in fine-grained complexity to produce multi-verifier signatures of work, which the consensus protocol picks up as a modular component. In short, one dependency from the earlier design has been swapped out. The stack of cryptographic assumptions is still standing.
What nobody can tell you yet
For permissionless consensus, the result is still theoretical, and the holes in the public material are specific enough to name.
No numerical corruption threshold has been stated for the 2026 protocol. No exact synchrony and communication conditions. No concrete costs. No benchmark performance. And nothing from the 2024 construction can be assumed to carry over merely because some of the same names sit on both papers.
The upshot: the evidence on hand supports neither comparisons with Bitcoin’s security model nor claims that this is deployment-ready. Any thread you see this week doing either is running ahead of the paper.
You can’t read it yet, and that matters
The Edinburgh record says the accepted manuscript remains embargoed until Aug. 20 at 00:00 BST. That is the conference’s closing day — a fairly narrow window between the talk and the text.
In the meantime, the entire claim, stripped of everything people will be tempted to bolt onto it: distributed samplers and a common random string take over from a recurring randomness beacon, while DDH, LWE and fine-grained proof-of-work assumptions remain in the construction.
Put Aug. 20 in the calendar, and go to the corruption threshold and the synchrony conditions first. Those two numbers will say more about whether the construction means anything practical than any abstract can.
