AI Ethics
Responsible AI, measured — not marketed.
Krex helps you make and check content with AI. Here's exactly where we stand on disclosure, consent, your data, and what we refuse to help with — written to match what the product actually does, not what sounds good.
We measure honestly, and publish what we find.
Our AI detector reports curves, not confidence theater. We ran it against a labeled evaluation set until we understood exactly where it breaks, then scoped our product claim to the data instead of the other way around. Every scan says plainly that a clean result does not prove a human wrote something — current AI told to write casually evades every detector in this category, ours included. The full write-up is public: how we measured our detector until it broke.
We won't help you evade detection or fake authenticity.
The Analyzer refuses to humanize AI text to beat detectors, fabricate citations, or manufacture reviews. Those are the exact misuses these tools exist to catch — building a back door around them would defeat the point. We also believe AI-assisted work should be disclosed where it matters, not hidden.
Your content is yours. Full stop.
Documents and scans live in your private history. You can delete them any time — trash, restore, or purge permanently. We don't use your content to train models, and we never turn it into a marketing example without asking. See the privacy policy for specifics.
We don't fabricate.
When the citation checker can't verify a reference against a real registry, it says so — it returns “not determinable,” it doesn't invent a source to look confident. When the evidence is thin, the detector withholds an accusation rather than manufacture one. Uncertainty is disclosed, never dressed up as certainty.
Likeness and voice need consent.
Avatars and Personal Clones are built from your own photo and voice, or from talent licensed for it. We don't clone a real person's face or voice without their consent, and we don't help you impersonate someone.
You decide; the tool assists.
Nothing auto-publishes without your approval. The Ad Copilot plans the whole ad — and shows you the exact credit cost — before it spends anything. A human stays in the loop on every render and every post.
Honest pricing.
Credits are model-weighted, and the cost of every render is shown before you confirm it. We don't bury the price or advertise a flat rate that isn't true.
What we won't build for you
Some requests we decline regardless of who's asking or why — the guardrails hold even when it costs us a sale:
- Evading AI detection, or “humanizing” text to beat detectors
- Fabricated citations or fake academic sources
- Impersonating a real person's face, voice, or identity
- Fake reviews, testimonials, or records presented as genuine
- Content that targets or harasses a private individual
This is a living document — as the tools grow, so does this. If we ever fall short of it, tell us.