Created from a single voice note with Agent Craft

Everyone spent this week talking about AI watermarking. Nobody's…

Everyone spent this week talking about AI watermarking. Nobody's asking the real question. Who else is doing this? Google has been watermarking Gemini's text for a while now, with something called SynthID. Over 100 billion pieces of content marked across their products. Anthropic just announced their own version. They're not the first here. They're the second. OpenAI hasn't shipped this into their text. They've researched it, they just haven't done it. But they have marked their images and audio using the same tech Google's using, SynthID, alongside NVIDIA, Elevenlabs, and a handful of other vendors. Here's where it gets strange. Every one of these marks is vendor-specific. Google detects Google. Anthropic detects Anthropic. There is no universal "was this AI" button, and there isn't going to be one. You basically have to already suspect who made something, and then check it yourself. Now the part that changes everything. The image generators that don't participate at all. Midjourney, Stability, the open models anyone can run on their own laptop. They're marking nothing, and nobody can make them. Which means the labeling is being done by the labs least likely to be the problem, and the tools built for actual deception stay invisible. I don't think that makes it pointless. A world where most synthetic content carries some trace is better than one that doesn't. But be honest about what this solves. It solves the accountability problem for the companies that already want to be accountable. That's not the same as solving it. So if a detector can only tell you "this is Claude" or "this is Gemini", is that transparency, or is it just a receipt? -- Repost it to your network and follow for more. James Goddard - Co-founder & VP of Engineering Powered by Agent Craft