As of this month, it’s law. The EU AI Act’s Article 50 requires labeling of AI-generated content starting August 2, 2026, and the EU’s Code of Practice on AI-generated content marking prescribes a multi-layer approach combining C2PA metadata, imperceptible watermarking, and centralized logging of generation and modification events. On the US side, California SB 942, effective January 2026, requires large AI providers to offer provenance detection tools to users. Watermarking isn’t a future consideration. It’s here.
I asked Claude.ai for guidance on what this will mean for individual contributors and organizational integrity. Here’s the breakdown, pros and cons.
Most of the watermarking conversation focuses on companies and policy. But the real test happens at the individual level, every time someone drafts an email, a proposal, or a deliverable with AI assistance.
The Upside
Watermarking makes disclosure the default instead of a judgment call. You no longer have to decide in the moment whether to mention AI assistance. The record does it for you.
It rewards people who genuinely add value on top of AI output. If a watermark shows what was AI-assisted, the editing, judgment, and domain expertise you layer on top becomes the visible differentiator, not the words on the page.
It protects you when the stakes are high. If you can demonstrate you can produce the same quality of work without AI, a watermark on your everyday output isn’t a liability. It’s proof you’re using a tool, not depending on one.
It creates a paper trail for accountability. If a decision or communication is later questioned, provenance data shows exactly what was generated versus what was human judgment.
The Friction
It exposes people who’ve quietly represented AI output as entirely their own thinking, which can be an uncomfortable reckoning for individual credibility, not just organizational policy.
It raises a real question for confidential and secure environments. If you can’t run sensitive material through AI tools at all, your skill has to hold up independently. Watermarking indirectly tests whether that’s true.
It can create unfair scrutiny. A watermark showing AI assistance on a first draft doesn’t capture the hours of human refinement that followed. People will need to get comfortable explaining that distinction.
It’s inconsistent across platforms right now, so contributors can’t yet count on it to fairly represent their actual process.
Where I Land
There’s nothing wrong with using AI, so long as you don’t take credit for it, and provided you can complete the work independently of AI if that’s a requirement, such as for confidential or secure communications. That’s the standard: not “did you use AI,” but “can you still do the work without it, and are you honest about which parts were yours.” Watermarking doesn’t create that standard. It just makes it visible.
At RCG Workgroup, this is the line I coach clients on: AI is a tool, not a crutch. A tool sharpens judgment you already have. A crutch replaces judgment you haven’t built. Watermarking is going to make that difference impossible to hide, so the work now is making sure it’s a distinction you’re proud to have visible.
Where do you draw that line in your own work?
