SecondMark position
The relevant question is not whether AI touched the code. It is whether the complete engineering process produced evidence strong enough to support the system's consequential claims.
The examination is bounded to defined claims, a versioned system, and evidence that can be traced to the state under review. Any material exclusion or uncertainty remains visible in the conclusion.
Core questions
What the examination needs to establish.
- 01
Can implementation provenance be reconstructed?
- 02
Do tests cover behavior rather than generated structure alone?
- 03
Where can machine-produced assumptions escape review?
- 04
Can the released system be tied to the qualified evidence?
Evidence model
Evidence is assembled around the claim—not the folder structure.
Architecture and system boundaries
Collected, attributed, challenged, and connected to the exact system boundary under examination.
Implementation and model provenance
Collected, attributed, challenged, and connected to the exact system boundary under examination.
Runtime behavior and operational history
Collected, attributed, challenged, and connected to the exact system boundary under examination.
Controls, qualifications, and unresolved risk
Collected, attributed, challenged, and connected to the exact system boundary under examination.
Intended outcome
A defensible view of whether the AI-engineered system, not merely its repository, supports the claims placed upon it.
Professional boundary
What an opinion does—and does not—mean.
It provides
A traceable independent conclusion on defined claims, grounded in the evidence and system state examined.
It does not provide
A guarantee that failure is impossible, a permanent certification, or a conclusion beyond the stated scope and validity conditions.