Acceptable use
A detector aimed at a person is a tool that can harm them. These limits are the ones we would want applied to a verdict about our own voice.
Do not do these things
- Do not make a Truthring result the sole basis for an adverse decision. Not a dismissal, not a rejection, not a disciplinary finding, not a prosecution. We publish a false positive rate because it is not zero, and the person on the wrong side of it cannot prove a negative.
- Do not use it to monitor people covertly. Employees, partners, family members, children. Surveillance is not a use case we support and accounts doing it will be closed.
- Do not submit audio you obtained unlawfully. Recording rules differ by jurisdiction. Obtaining a recording illegally does not become acceptable because of what the analysis returns.
- Do not present a verdict as proof. In publication, in a filing, in an accusation. Report it as a probability, name the model version, and state the error rate. Misrepresenting our output as certainty is a breach of these terms as well as of the underlying facts.
- Do not use results to train a system that evades detection. Systematic probing to find and exploit our failure conditions.
- Do not resell or rebrand the output without a written agreement, and never in a way that hides the model version or the error rate from the end reader.
- Do not scrape or automate the web interface. There is an API; it is priced so this is unnecessary.
The rule underneath all of them. If a person could be harmed by the verdict, they are entitled to know a probabilistic tool was used, what its error rate is, and that a human made the decision. Anything that removes one of those three is outside acceptable use, whether or not it is listed above.
What happens when this is broken
[VERIFY: your actual escalation. Typically: contact first for something that looks like a misunderstanding, suspension for a clear breach, immediate closure for surveillance or for misrepresenting output as proof in a proceeding. State it plainly and then follow it.]
Reports of misuse: abuse@aivoicedetctor.com
Reviewed