Attribute-based encryption has been around since 2004, when Amit Sahai and Brent Waters proposed it. The idea is elegant. Instead of encrypting a file for named people, you encrypt it for a policy (say, finance AND Europe AND manager), and anyone whose keys carry matching attributes can open it. It stayed mostly academic, and the quantum threat added a second problem, since the practical schemes ran on pairing-based math that a quantum computer would break.
Saltgrain and the NTT Research Cryptography and Information Security (CIS) Lab now say they have the first practical large-universe post-quantum ABE. It is headed for Saltgrain, the data-centric security platform for sovereign data that came out of NTT Research’s Scale Academy incubator in April. Integration testing has started. There’s no availability date yet.
What large-universe means
Many ABE schemes make you list every possible attribute when the system is set up. Large-universe schemes drop that requirement, so an organization can add a new label (a new department, a new region, a new project) whenever it needs one. A multi-authority version lets separate organizations or departments issue credentials independently, which matters once data crosses company or national lines.
The schemes are built on Learning With Errors and its ring variant, the same lattice family behind NIST’s post-quantum standards. The ABE schemes themselves are not NIST-standardized. The team built both key-policy and ciphertext-policy versions.
The speed numbers
Benchmarks ran on an Apple M4 Pro MacBook Pro, encrypting a 256-bit AES key under AND policies of one to four attributes:
- Decryption: 5.6 to 22.5 ms for key-policy, 7.5 to 33.5 ms for ciphertext-policy
- Encryption: 195 to 545 ms for key-policy
- The Ring-LWE versions pack 2,048 message bits per ciphertext, about 2,048 times more compact per protected bit than the one-bit versions
- Roughly 128-bit security for the key-policy scheme
The catch is size. Keys and ciphertexts run to several megabytes or more. The companies say plainly these are research benchmarks, not product claims.
“On an ordinary laptop, a user can unlock a protected file in a few hundredths of a second,” said Yannis Rouselakis, senior cryptographic developer at the NTT Research CIS Lab and Saltgrain. That, he argues, is where it stops being theory and starts being a product.
Why now
NIST finalized its post-quantum standards in 2024, and Executive Order 14412 sets federal deadlines for 2030 and 2031, with agency migration plans due now. The standing argument is harvest now, decrypt later: data stolen today gets read once a capable quantum computer exists. Saltgrain adds an AI angle. Companies want agents working through big document sets full of private data, and encryption that carries its access policy with the data is one way to give an agent the right slice and nothing more.
The paper, “Large-Universe (Multi-Authority) ABE from LWE” by Pratish Datta, Yannis Rouselakis, Junichi Tomida and Nikhil Vanjani, is being presented at ACM CCS 2026 in The Hague this week.