Maintenance
Status
How current the records are. Each record has a review interval, usually 90 days, and is flagged when a review is overdue.
104 records
104 published · 0 in draft
0 overdue reviews
Records past their review interval.
0 disputed
Assessments under formal dispute.
5 with open critical flaws
A published flaw that defeats the main claim under the stated threat model.
Open critical flaws
- R3Model identity attestation⚠
- R2On-chip telemetry from timing, memory and performance counters⚠
- R3TEE remote attestation for AI workloads⚠
- R2Training-transcript verification (proof-of-learning)⚠
- R3Tinfoil model identity (Modelwrap)⚠
Low-confidence assessments
Readiness assessments held with low confidence. These most need checking by people who know the work.
- R2Bandwidth limits and compartmentalization
- R2Bounding unexplained information in outputs
- R3Deterministic and bit-exact inference
- R1Proofs of useful work for capacity accounting
- R2Safeguard attestation
- R3Sampled inference recomputation
- R2Attestable Audits
- R1Attestable zero-knowledge inference prover
- R2Batch-invariant inference kernels (Thinking Machines)
- R3Pearl proof-of-useful-work blockchain
- R1RAND secure inference data center (SIDC) design
- R3TOPLOC
- R3Verde and RepOps (Gensyn)