Creators Blast AI Watermarks as 'Arsonist's Fire Extinguisher'

Leading AI developers, including OpenAI and Google, are implementing text watermarking systems to identify AI-generated content. While OpenAI has developed a system, it's not yet publicly available for ChatGPT, and Google applies its own SynthID Text to Gemini output. Both companies, along with Apple, use standard C2PA watermarks for images, with Google adding a proprietary one to resist manipulation.

However, this move has sparked significant criticism from writers and users. Detractors label detection systems as 'antidotes owned and sold by the same companies profiting from the poison' or 'fire extinguishers designed by arsonists,' likening the AI detection industry to a protection racket. Users of Anthropic's Claude, in particular, have voiced strong opposition, arguing that watermarks will diminish output quality, especially for code, and may discourage the use of AI for refining human-made work due to the risk of flagging.

These critics fear a scenario where legitimate users who used AI for minor edits could be penalized, while malicious actors could potentially bypass detection to pass off AI-generated propaganda as human-created. Anthropic acknowledges that its watermarking tool might flag human-made work that has been processed by AI, indicating it only proves AI involvement, not full AI authorship.