# AI-Generated Content Detection: Identifying Synthetic Text, Deepfake Images, and AI-Authored Media
Status: public
Confidence: medium (0.865) (verified)
Last verified: 2026-05-28
Generation: ai_structured

## TL;DR
AI-generated content detection is not a single solved classifier. It includes post-hoc text or media detectors, generation-time watermarking, and provenance systems that record how content was created or edited.

## Core Explanation
Post-hoc detection tries to infer whether text or media came from a generative model after the fact. DetectGPT is an example for text: it uses probability-curvature behavior of language models rather than a separately trained detector. Watermarking works earlier in the pipeline by changing generation so outputs carry a statistical signal. Provenance standards such as C2PA take a different route: they do not prove that all unlabeled content is fake, but they can attach signed creation and edit history to content that participates in the standard.

## Further Reading

- [DetectGPT](https://arxiv.org/abs/2301.11305)
- [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226)
- [C2PA Technical Specification](https://spec.c2pa.org/specifications/specifications/2.3/specs/C2PA_Specification.html)