What it is
This research studies how computation can be protected, checked, or proven so that sensitive workflows are not reduced to plain-data processing.
How it is researched
The lab models protected computation paths, proof profiles, witness checks, and prototype multiplication workloads, then separates what is measured from what still needs external review.
Why it matters
It gives the portfolio a privacy and verification layer for AI/security workflows where the result must be trusted without exposing every internal value.
Evidence
- Step7 script path: 19/19 internal scripts passed.
- Scoped P128 proving profile: 1.06 ms vs snarkjs 1,945 ms, 1,835x under the named profile.
- FHE/vFHE Hom.Mult prototype: 2.359 ms vs TFHE-rs 149 ms, 63.16x prototype wall-time.
- Bootstrap witness path: 512/512 internal checks.