Comment by bawana

If openAI and Anthropic have found ways to watermark text as 'AI generated' then this is a communications channel. AI agents can learn this algorithm and use this channel to communicate and we will never know.