ChatGPT
Visit ChatGPTAmbient Clinical Scribe · model chatgpt-gpt-4o · tested Aug 25, 2026 · published Aug 25, 2026
Did not meet certification thresholds on this run.
VerifiedConnected mode — the certification body administered the run and captured the evidence directly from the vendor’s system. Evidence channel: sequential API collection.
Fabrication rate
1.1%
95% CI 0.4–2.9%
Tier A target < 1.0%
Severity-weighted omission
4.5%
95% CI 1.8–10.7%
Tier C ceiling < 25%
Unsupported inference
7.4%
95% CI 5.1–10.7%
Weighted in composite
Cross-subject containment
Every test packet carries unique canary details. A pass means no information from one subject's packet ever appeared in another subject's output — a single confirmed leak fails the run outright.
Evaluation coverage
Scope of the certification run behind this attestation.
Test cases
15
Difficulty tiers 1–4
Tracks
3
Cardiology · Primary Care · Psychiatry
Facts checked
353
349 assertions graded
Judge–reviewer agreement
100%
On adjudicated claims
Claim classification
Certification scale
Methodology
Controlled synthetic cases; atomic claim grounding vs. the source packet; per-run canary registry for cross-subject containment testing; reviewer-adjudicated low-confidence, fabrication, containment, and severe-omission claims; severity-weighted tiering. Rates carry 95% Wilson confidence intervals, and tiers gate on the conservative (upper) bound of the interval rather than the point estimate.
Attestation history · Ambient Clinical Scribe
| Published | Tier | Assurance | Bench | Model | Fabrication | Composite | Status |
|---|---|---|---|---|---|---|---|
| Aug 25, 2026 | — | Verified | v1 | chatgpt-gpt-4o | 1.1% | 0.97 | active |
All evaluations use synthetic cases. Certification reflects measured output accuracy on the FairAttest bench at the test date.
Verify this attestation independently via the machine-readable export.