Consent-first AI: why nothing should ship silently

AI translation is fast enough to be dangerous. When suggestions are applied automatically, errors propagate to production before anyone notices - and there’s no record of what changed or why.

Consent-first AI flips the default. The AI suggests; a human approves. Nothing is overwritten silently.

Every decision is traceable

In StringsOS, accepted suggestions and human overrides are preserved in translation history so reviewers can understand how the current value was chosen.

That traceability keeps AI-assisted localization reviewable without claiming that an unfinished enterprise audit system already exists.