Independent benchmark · verifiable inference
57%of audited providers can't prove which model you were served.ⓘ
0% · model checkableblind trust · 100%
7 audited · 3 let you verify the model · 4 you can't · 0 caught serving a wrong model · cycle 39 · 2026-08-10 08:03
7 providers · click a row for proof
RedPill (Phala)attested · tdx · h100
1 verifier
1 verifier90intel-rooted seal · tcb ok
0
passreproduced
Venice.aiattested · intel-tdx · nvidia-cc · phala1 verifier
1 verifier90intel-rooted + hopper seal · tcb ok
0
passas submitted
NEAR AIattested · tdx · h1001 verifier
1 verifier72intel-rooted seal · tcb OutOfDate
0
partialreproduced
PPQ.AIattested · sev-snp · proxy1 verifier
1 verifier60no reference
0
partialas submitted
NanoGPTattested · intel-tdx · nvidia-cc · resold1 verifier
1 verifier40tdx seal · dcap pending
0
partialunverified
Chutesattested · intel-tdx · nvidia-cc · opaque1 verifier
1 verifier0no reference
0
errorunverified
EigenAI (EigenCloud)attested · intel-tdx · eigen · inside-only · proxy
—no verifiable seal
—
unknowninside-only
The frontier
open challenge · no submissions yetThe best detector so far catches 8 of 14 swapped providers without falsely accusing more than one honest one. Beat it and yours takes over.