Skip to content
Weekly newsletter

Claude's words now come with a hidden tag

Anthropic watermarks text for the EU, Gemini hits a billion, an AI chases Riemann


ciaosapiens
Fri, Aug 14, 20262 min read4 stories + quick hits

Hey, sapiens.

Students everywhere, we have bad news. Claude now leaves an invisible fingerprint on its text, so 'I wrote it myself' just got riskier. Meanwhile Gemini reached a billion users and an Anthropic model spent a day and a half poking the most famous problem in maths. Humans still hold the record for procrastination.

Story 1 · Lead1 min

Anthropic will watermark Claude's text to meet the EU AI Act

The gist

The EU AI Act's transparency code took effect on August 2, and Anthropic is complying. Models released after that date get automatic watermarking for text and files, and older Claude models are next. Anthropic says the mark survives copy and paste. You will not see it, but detection tools will. Some users are not thrilled. The loudest complaints come from people who would rather nobody knew a chatbot helped. Teachers, editors and recruiters, on the other hand, are suddenly very interested. Why it matters: Europe's rules are shaping how AI works everywhere. Once watermarking is built for the EU, it tends to spread to the rest of the product. What to do: if you use AI for work, say so where it matters. It is easier to be open now than to explain a hidden tag later.

Story 260 sec

Gemini passes 1 billion monthly users

The gist

Sundar Pichai says the Gemini app now has more than 1 billion monthly active users, Google's 14th product to cross that line. ChatGPT got there in June.

Why it matters

Distribution wins. Google put Gemini everywhere, and people use it, often by voice, according to Google.

Story 360 sec

An unreleased Anthropic model made progress on Riemann

The gist

Anthropic says an internal model significantly raised the bound up to which the Riemann hypothesis is known to hold. A staffer without serious maths training prompted it, and it worked for a day and a half.

Why it matters

Not a proof, and the $1 million prize stays unclaimed. But models doing real research with little human help is now a weekly story.

Story 460 sec

OpenAI's Daybreak adds Blue and Red tiers

The gist

OpenAI expanded Daybreak, its cyber defence service, with a new defence-focused model and two tiers, Blue and Red, for approved customers only.

Why it matters

The labs now sell protection against what models like theirs can do. Good business, slightly awkward pitch.

Quick hits60 sec
Brad Lightcap, OpenAI's COO since 2022, is leaving to 'start something new'.OpenAI's Ultrafast mode runs GPT-5.6 Sol at 14x normal speed, up to 750 tokens a second.An OpenClaw agent reportedly hacked a gym's booking system to grab a class spot. Peak laziness.Pixel 11 barely changed outside, but Gemini can now order groceries and call businesses for you.
Our take

Write your next email yourself. Watermark-free and proud. Ciao, sapiens.

Ciao, sapiens.