ArtifactNet provides forensic indicators, not conclusive legal proof. Verdicts are statistical estimates based on machine learning models and are subject to error. Do not use ArtifactNet as the sole basis for legal action, public accusation, contract enforcement, or copyright dispute resolution.
ArtifactNet analyzes audio recordings and estimates the probability that the recording contains AI-generated content. Each result reports that probability, the decision threshold it was compared with, and one of three labels:
ai): the score is at or above the decision
threshold.real): the score is below the decision
threshold. This does not establish that a person performed or produced the
recording, or that no AI tools were used.partial): some parts of the recording score
above the threshold and some below.A label describes what the detector found in the audio, not who made the recording.
Audio recompression (mp3 / aac / opus) can shift verdicts. We continuously improve codec invariance, but heavily processed mp3 streams (e.g., low-bitrate YouTube re-encodings) may produce false positives or false negatives.
ArtifactNet is trained on commercial generators known as of the model release date. Newer or private models may produce outputs that deviate from learned distributions. We update the model regularly to address this.
The following categories have higher error rates:
Even at 99% accuracy on a benchmark, 1 in 100 verdicts will be wrong. Treat each verdict as one input among many in your decision process, not as ground truth.
If you believe ArtifactNet has incorrectly classified your work, you may request a manual review by emailing disputes@intrect.io with:
We aim to respond within 5 business days.
The Service is provided "AS IS" without warranties of any kind, express or implied, including but not limited to merchantability, fitness for a particular purpose, or non-infringement. We do not warrant that verdicts are accurate, complete, or suitable for any particular use.