The only privacy that survives the next 20 years.
Every privacy system built on elliptic curves has an expiration date. H33's privacy stack — FHE for computation on encrypted data, STARK-based ZK for proofs, and triple-family post-quantum signatures for receipts — has none.
No PLONK. No Groth16. No BLS. No KZG. No pairings. The math is post-quantum end to end, and the proof of it is a portable receipt you can verify in the browser without us.
Compute on encrypted data
The H33 fully homomorphic encryption stack — three engines, one verifier, post-quantum end to end.
/fhe/Production deployment surface for FHE compute — runtimes, key management, and the verifier contract.
/fhe-platform/TFHE-256, CKKS-256, TFHE-Bootstrap-256. NIST L5 parameter sets across the FHE family.
/h33-256/Exact-integer FHE for finite-precision workloads. Equality, comparison, lookup at 128-bit security.
/h33-128/Approximate FHE for floating-point computation — encrypted ML, encrypted statistics, encrypted analytics.
/h33-ckks/Gate-level FHE for arbitrary Boolean circuits. The substrate beneath encrypted-program execution.
/tfhe/Programmable bootstrapping for unbounded-depth FHE circuits.
/tfhe-bootstrap/The category boundary: data stays encrypted through the computation, decrypted only by the data owner.
/compute-on-encrypted-data/Technical architecture for the FHE stack — runtimes, key management, parameter sets, and the verifier contract.
/docs/encrypted-compute-architecture/Compute on the sensitive blob. Tamper one byte. The verifier rejects the result.
The receipt H33 emits at the end of an FHE computation contains the same triple-family signatures as a session receipt. Modify one byte of the encrypted output or the supporting metadata, and the verifier rejects under all three families. This is the buyer demo: a real claim about a real dataset, replayed under tamper.
Verticals
Query encrypted indexes without decryption. Database stays encrypted; the query stays encrypted; the result decrypts only on the client.
Run inference on encrypted inputs. The model never sees the prompt; the prompter never sees the weights.
The receipt for an FHE computation. Triple-family PQ signatures over the encrypted input, the encrypted output, and the circuit.
Beyond encrypt-at-rest and encrypt-in-transit — the third leg: encrypt during computation.
For banks, governments, healthcare networks — privacy that satisfies the regulator and the threat model simultaneously.
GDPR, HIPAA, CCPA — the cryptographic primitives that make compliance evidence portable.
DLP without decryption: detect-and-prevent on encrypted streams.
End-to-end pipelines where data stays encrypted from ingest to decision.
Why post-quantum, not "blockchain privacy"
Why curve-based privacy systems have an expiration date and STARK-based ones do not.
The attack model that makes today's encrypted traffic tomorrow's leaked dataset.
What changes when the FHE provider has to disappear and the receipts still verify.
The exact guarantees the substrate provides — substrate vs application boundary.
The whole PQ stack — KEM, signatures, FHE, ZK — at a single glance.
NIST L5 across the board. Parameter choices and the security argument for each.
Where Privacy sits in the stack
The Privacy primitives feed the same gate as everything else. An FHE computation, an encrypted-AI inference, or a sensitive-data query all produce inputs that flow through H33-Root authority bindings, get evaluated by the HATS gate, and exit as a portable .h33pqv.json receipt that anyone can verify without the provider.