Can one approved workload use a controlled local endpoint with the required operational controls?
- Duration
- 6–10 weeks after gate
- Availability
- Gated design-partner candidate
Scope and terms
Test one approved local-model workload behind a controlled endpoint with explicit identity, capacity, audit, restore, and rollback behavior.
One approved workload and supported configuration. No OpenAI-wide compatibility, general availability, production SLA, or customer-operability claim before the gate passes.
Can a second operator reproduce one bounded data-to-evidence workflow?
- Duration
- 8–16 weeks after gate
- Availability
- Gated co-development candidate
Scope and terms
Make one local AI training workflow reproducible from approved data to a versioned checkpoint, evaluation, and evidence pack—then prove a second operator can run it.
One registered data-to-evidence workflow, environment, operator cohort, model/profile, rights set, and acceptance protocol. Not a general training platform.
Can a repeated release-evidence process become a reviewer-operable workflow?
- Duration
- 6–10 weeks after gate
- Availability
- Gated design-partner candidate
Scope and terms
Turn one repeated AI release-evidence process into a controlled path from claim to artifact to test to approval—then prove a second reviewer can follow it.
One repeated release-evidence process, one minimum schema, file or bounded API-export integration, and one second-reviewer workflow. Not a generic governance or accreditation platform.