The DeepSensi Standard · DSS
Aviation has DO-178C.
Nuclear has IEC 61508.
Medicine now has DSS.
An aircraft engine must demonstrate a failure rate below 10⁻⁹ per flight hour before it enters service. A clinical AI can currently be deployed in a hospital with no quantitative reliability requirement at all. The DeepSensi Standard closes that gap: the first certification framework that assigns clinical AI a measurable, auditable hallucination-probability bound — assessed with the same Fault Tree methodology the safety-critical industries have used for half a century.
Five pillars — each independently auditable
Four levels — explicit bounds
| Level | Hallucination bound | Reference point¹ | Intended deployment |
|---|---|---|---|
| DSS Bronze | < 10⁻³ | SIL 3 target | Health information, wellness, screening |
| DSS Silver | < 10⁻⁵ | SIL 4 target | Primary-care decision support, telemedicine |
| DSS Gold | < 10⁻⁷ | ≈ DAL B | Hospitals, specialist clinics, regulatory submissions |
| DSS Platinum | < 10⁻⁹ | ≈ DAL A | National health systems, military, sovereign AI |
¹ IEC 61508 demand-mode numerical targets; the SIL scale saturates above Silver, so comparison continues via DO-178C Design Assurance Levels (per-flight-hour — reference points, not equivalences). Bronze and Silver may be self-assessed; Gold and Platinum require third-party audit, continuous barrier telemetry, and 24-month recertification.
How certification works
An applicant documents its verification barriers and their independence, assigns per-barrier failure probabilities with supporting evidence, computes the system-level bound via Fault Tree Analysis with a justified common-cause β-factor, and multiplies by worst-case degradation. The worst-case bound determines the certified level. The methodology is fully specified in DSS-001; the companion mathematics in WP-001.
DeepSensi's own platform implements the complete Platinum architecture, with a conservatively certified bound of 3.23 × 10⁻⁶ worst-case. We publish that distinction deliberately — a standard that flatters its own author is not a standard.
"A safety standard behind a paywall is not a standard. It is a product."
For hospitals: require a DSS level in your next AI procurement. For insurers: price coverage against a certified bound instead of excluding AI outright. For regulators: the framework maps directly onto EU AI Act Articles 9, 10, 14, 15, and 17. For vendors: certify against it — freely.