Brightfold

Brightfold — District-wide AI prompt/response decision logs: every student and staff request routed through the Gateway policy layer, paired with the guardrail verdict (allow / redact / block), the target AI tool, PII categories detected, safety-escalation (mandated-reporting / crisis-signal) flags, and immutable audit-trail metadata.. Small company, unusually dense corpus: an early-stage single-line K-12 edtech vendor by headcount and district count, yet each deployed district yields millions of policy-labelled AI conversations per year. The value is density and label quality (verdict-labell…

Brightfold offers (Alternative, Sentiment) — Est. tens of millions of prompt/response pairs per year. Basis: the page states an average district sees roughly 3,000 ed-tech tools a year through the gateway; one 20,000-student district at ~5 AI requests per user per day over a 180-day school year is about 18M decisions, and each stored decision carries both the prompt and the model response..

Training and evaluating minor-safety and content-moderation classifiers; jailbreak / prompt-injection red-team corpora under real attack traffic; PII detection and redaction models tuned to child-authored text; escalation and crisis-language detection for school-safety systems; ed-tech AI-tool adoption telemetry for education market intelligence; preference data on which model outputs a district's policy layer accepts.

The data is with 2 years of history.

Coverage spans US; Application Software; alternative, sentiment; equities.