01Why we publish these principles
HUMXN supplies human-made data used to build AI systems. The way that data is sourced affects creators, the people who appear in it, and everyone who uses the models trained on it. These principles describe the commitments we hold ourselves to and how they show up in the product.
02Consent first
Every work carries an explicit AI-training permission chosen by its contributor, bound to the work itself in its provenance record. Works showing people are only offered for training when consent or a release is on file. Contributors can see how their work is licensed.
03Fair pay
Contributors and experts are paid for their work. Experts see the rate for a brief before they accept it, and payouts are made through Stripe after verification. We aim to make compensation clear, predictable and proportionate to the skill involved.
04Transparency about human origin
Each work records how it was made: human-created, AI-assisted or AI-generated. Machine-readable signals such as generator signatures, IPTC digital-source-type fields and C2PA generation records override a human-created declaration. Labels record their source, so buyers can tell expert-reviewed labels apart from creator statements and AI suggestions.
05Privacy by design
Documents and tables are scanned for personal data, which is masked before delivery. Expert identities are not disclosed to buyers without consent. Our public website measures visits without cookies and without storing IP addresses.
06Safety screening
Uploads are screened against hash block-lists and a safety classifier, and we have zero tolerance for child sexual abuse material. AI tools assist our reviewers; people make the decisions that matter, such as accepting experts and resolving conflicts.
07No data from minors
Contributors and experts must be adults. We do not knowingly source data from or about children, and works depicting minors are not offered for AI training.
08Honest about the limits of technology
Provenance tools are powerful but not magic, and we try to be precise about what they prove:
- Crawler rules such as robots.txt and AI-training opt-out signals are declarations of the owner's wishes; they do not technically prevent copying. We record and respect them, but they are not a guarantee.
- AI-content detection is probabilistic. A work that shows no generator signals may still have been made with AI, which is why we combine signals with declarations, review and a signed record of who said what.
- Signatures and timestamps prove that a record existed unchanged at a point in time; they do not by themselves prove who authored the underlying work.
- Forensic marks support tracing and enforcement but can be degraded by heavy editing.
09Accountability
Every change to a work is written to a signed, hash-chained audit trail, and every delivery comes with a signed receipt, so our claims can be checked rather than taken on trust. We review these principles as the technology, the law and our understanding change.
10How to raise a concern
If you believe your work, likeness or personal data is in our registry or a dataset without proper consent, or you have any concern about how we apply these principles, email hello@humxn.io. For privacy requests, use privacy@humxn.io. We will acknowledge your message, look into it, and where content was included without the required rights or consent, withhold it from future datasets while we resolve it.