bleepingcomputer[.]com/news/artificial-intelligence/how-anthropic-plans-to-watermark-claudes-ai-generated-text/
There's probably no 100% accurate way to tell if something is AI generated, though I've read that with video you can come close, if you want to take a deep enough dive, going frame by frame. Watermarking an image invisibly generally involves altering some unobtrusive pixels, while for text it involves slightly altering the choice of words used when there is a choice, e.g., big vs. large. A software AI detector made by a company like Anthropic would know what patterns to look for to detect text generated by one of its models, but likely wouldn't have a clue if the text came from OpenAI. I'd also guess that students for example would be smart enough to replace adjectives to disrupt that pattern [assuming they cared].