policyAug 13, 2026 · EU AI Act, Article 50 Transparency Code · 3 min read

AI writing now carries a watermark, but it can't prove who did the thinking

Morning Byte · by Keith Andrade

Comments

A woman in a dark blazer with a lanyard sits at a table by a window, a red pencil in her hand, lifting the corner of a printed document. Beneath the raised page a dense blue grid of glowing points is revealed, as though embedded in the paper itself. Outside, European Union flags fly in front of the Berlaymont building at sunrise. A colleague works at a desk behind her.

The three takeaways

EU transparency rules took effect August 2: AI-generated text must be machine-identifiable.

Anthropic's watermark proves a document passed through Claude, not who wrote its ideas.

Assume AI attribution is visible. Lead with your judgment.

Source

EU AI Act, Article 50 Transparency Code · Anthropic policy statement

Morning Byte · weekly digest

One email a week. The stories that mattered.

What changed

On August 2, the EU's transparency rules for AI-generated content took effect. The short version: if a machine wrote it, a machine has to be able to tell. AI companies now mark the text their models produce, and Anthropic, the company behind Claude, has started watermarking accordingly.

The detail most coverage misses: the watermark says a document was processed by AI. It says nothing about whose thinking is inside it.

Aug 2, 2026

The date AI-generated text became machine-identifiable across the EU.

When the transparency rules took effect.Source: EU AI Act, Article 50 Transparency Code

Aug 2024AI Act in force
The EU's framework becomes law.
Feb 2025First bans
Prohibited uses switch off.
Aug 2025Model rules
Obligations reach the AI companies.
Aug 2, 2026Transparency
AI text must be identifiable.

How the EU got here: the AI Act's rollout.Source: EU AI Act

Passed through AIestablished

Who wrote the ideasnot established

What the watermark can and cannot establish.Source: Anthropic policy statement · Hatched = not established

Why it matters

If you use AI at work for drafts, emails, or reports, the question stops being ‘can anyone tell?’ and becomes ‘what happens when they can?’ The honest answer: the same thing that always happened. Work gets judged on whether it's right and useful. The watermark just removes the guessing game.

What to watch

How employers treat watermarked documents. If a flagged draft gets read on its merits, this becomes a non-event. If it becomes a filter, the people who can show their judgment on top of the tool win.