Illustrative image: encryption keeps data locked while it is processed. Image: Wikimedia Commons, public domain.
📌 Key Highlights
- What: TUTTI, a new parallel bootstrapping procedure for the CKKS fully homomorphic encryption scheme, posted to arXiv on 9 November 2026.
- Headline number: a single-message bootstrap in 2.4 ms on an Apple M5 Pro laptop (14 threads) — under the 3 ms in the paper's title — against 4.5 ms for TFHE-rs on the same machine.
- At scale: 4,096 real-number messages refreshed in 46–83.4 ms on 192 threads, up to 3× faster than SHIP, the previous state of the art.
- Why it matters: it closes the long-standing speed gap between CKKS and lighter schemes, so hospitals, banks and AI services could analyse data without ever decrypting it.
- Caveat: this is a preprint, not yet peer-reviewed, and differences in key sizes and security assumptions between CKKS and TFHE remain.
What happened?
On 9 November 2026, researchers Robin Köstler and Antonio Guimarães posted a paper titled "TUTTI: Parallel CKKS Bootstrapping in under 3 ms on CPU" to the arXiv preprint server. It tackles one of the oldest problems in cryptography: fully homomorphic encryption (FHE) lets you compute on data while it stays encrypted — a hospital could let an AI model analyse patient records, or a bank could run fraud checks, without anyone ever seeing the raw data. The catch has always been speed. Every calculation on encrypted data adds "noise", and the clean-up step that removes it — called bootstrapping — used to take seconds or even minutes, making FHE useless for anything real-time.
TUTTI changes that trade-off. It builds on an earlier method called SPRU, splitting the work across many CPU threads at once, with a refined security analysis of the block-binary secret keys the scheme relies on. The result is the fastest CKKS bootstrapping ever reported on a CPU — and, for a single message, faster than TFHE, the lighter scheme that has traditionally owned the low-latency crown. It joins a strong run of algorithm breakthroughs this month, including SwarmEvolve, which uses silent AI agent swarms to discover better algorithms in minutes.
TUTTI CKKS bootstrapping under 3ms: what the paper actually shows
The headline needs one careful reading. The paper's title promises "under 3 ms", and that figure comes from the laptop test: on an Apple M5 Pro, TUTTI refreshed a single encrypted message in 2.4 milliseconds using 14 threads, while TFHE-rs needed 4.5 ms on the same machine. On a server CPU, the single-message time was 4.6 ms on 32 threads — still faster than TFHE-rs at 7.3 ms, but above 3 ms. Both numbers matter, and both are in the paper's abstract, so here they are side by side:
| Test | TUTTI | Best previous / rival |
|---|---|---|
| Single message, server CPU (32 threads) | 4.6 ms | TFHE-rs: 7.3 ms (4-bit bootstrapping) |
| Single message, Apple M5 Pro laptop (14 threads) | 2.4 ms | TFHE-rs: 4.5 ms |
| 256 messages, server CPU (32 threads) | 15–34 ms | Gains immediately even at small batches |
| 4,096 messages, server CPU (192 threads) | 46–83.4 ms | SHIP: up to 3× slower on the same machine |
The 4,096-message result is the one that matters most for real workloads: CKKS packs thousands of real numbers into a single ciphertext, so refreshing 4,096 values in under a tenth of a second — while restoring 1 to 9 multiplicative levels for further computation — is what turns FHE from a lab demo into something a cloud service could actually run. Unlike SHIP, the previous record-holder, TUTTI also gains speed immediately when bootstrapping fewer messages, instead of only paying off at huge batch sizes.
Illustrative image: a homomorphic encryption setup — data stays encrypted through the whole computation. Image: Wikimedia Commons, CC-Zero.
Why it matters for India readers
India is writing the rules for exactly this kind of technology right now. Under the Digital Personal Data Protection Act, hospitals, fintechs and government services must protect personal data even while they use it — and "compute without seeing the data" is the strongest version of that promise. A fast CKKS bootstrap means an Indian health-tech startup could run an AI diagnostic model on encrypted patient scans, or a bank could score encrypted transactions for fraud, without the raw data ever leaving the owner's control.
There is a practical angle too. TUTTI runs on ordinary CPUs — a laptop chip, not a data-centre GPU farm — which lowers the cost of entry for Indian developers working with open FHE libraries such as OpenFHE, and matters for privacy-preserving AI on phones and edge devices. None of this is shipping in a product yet — the paper is a preprint, and integrating it into production libraries is the next step — but the direction is clear: encrypted computing is getting fast enough to be ordinary.
Should you care?
✅ Yes, if you build with sensitive data. If you work in health-tech, fintech or cloud AI, TUTTI is a genuine milestone: the speed gap that kept CKKS out of real-time use is effectively closed on multi-threaded CPUs. Follow the paper's integration into OpenFHE and try the benchmarks yourself before betting a product on them. ⏳ Wait, if you just want the feature. This is research, not a release — no app or service offers TUTTI-speed encrypted computing today, and the authors themselves note that key sizes and security assumptions still differ between CKKS and TFHE.
FAQs
What is fully homomorphic encryption?
Fully homomorphic encryption (FHE) is a form of encryption that lets you add, multiply and run computations on encrypted data without decrypting it first. The result, once decrypted, matches what you would get from computing on the raw data — so a cloud server can process your information without ever seeing it.
What is CKKS bootstrapping?
CKKS is an FHE scheme designed for real numbers, created by Cheon, Kim, Kim and Song. Every operation on CKKS-encrypted data adds noise, and bootstrapping is the refresh step that cleans the noise away and restores room for more computation. It has historically been the slowest part of the whole scheme, often taking seconds.
How fast is TUTTI bootstrapping?
For a single message, 2.4 ms on an Apple M5 Pro laptop and 4.6 ms on a server CPU — both faster than TFHE-rs on the same machines. For 4,096 messages at once, 46–83.4 ms, up to three times faster than SHIP, the previous state of the art.
Is TUTTI peer-reviewed yet?
No. The paper was posted to arXiv on 9 November 2026 as a preprint, so its results have not yet been through formal peer review. The benchmarks are detailed and reproducible, but independent confirmation — and integration into libraries like OpenFHE — will be the real test.
Takeaway: TUTTI does not make encryption unbreakable — FHE was already that — it makes it fast. A 2.4 ms refresh on a laptop chip turns computing on encrypted data from a research curiosity into a practical tool, and the next thing to watch is how quickly it lands in the open-source libraries developers actually use. If you work with sensitive data, read the preprint linked below and check whether your stack could use it — our explainer on the Triangular Log-Det PDHG algorithm shows how the same "faster maths, same guarantees" pattern is playing out across optimisation research.
Sources
- arXiv:2611.12854 — TUTTI: Parallel CKKS Bootstrapping in under 3 ms on CPU (Köstler & Guimarães, submitted 9 Nov 2026)
- Striding Tech — Unbreakable Encryption, Real-Time Speed: A Breakthrough in Privacy-Preserving Computing (10 Nov 2026)
- Duality Technologies — FHE Bootstrapping: What It Is, How It Works, and Which Scheme to Use
Comments
Post a Comment