Definition
Machine-learning systems that classify text or images as human-authored, AI-generated, or mixed/AI-assisted — without relying solely on watermarks or metadata.
Key Points
- Hard cases: short text, AI-assisted editing, ai-humanizer tools, mixed human-AI passages
- Accuracy claims are historically brittle across models — treat vendor benchmarks cautiously
- Used by platforms, journalists, academia, and consumers fighting ai-slop
Detector accuracy claims (e.g. >99%) are vendor/internal; false positives and humanizer evasion remain contested industry-wide.
Related
- ai-generated-content-detection
- pangram
- ai-slop
- ai-humanizer
- content-moderation
- synthetic-mirror-training