Anthropic watermarks every Claude output; builders race to strip it

The mark rides generated text and signs files with C2PA across chat, API and Code; meanwhile MS Research's AutoGen lab folds and APIs churn 60+ times in a week.

Nowline NOV 9 12:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Claude now watermarks what it writes — by default

    Newer Claude models weave an imperceptible mark into generated text across the chatbot, API and Claude Code, and sign PNG/JPG/SVG files with cryptographic C2PA provenance — with no documented opt-out, and the mark survives copy-paste and light edits. Exact code and single-answer factual replies are skipped, so your prose is tagged while a tight code block likely isn't; open-source removers are already on GitHub while Anthropic's detector API stays 'planned.'

  • Microsoft Research's agent lab, AI Frontiers, is done

    John Langford says the ~50-person AI Frontiers group — the team behind AutoGen and the Fara computer-use agent — was folded into scaling-focused Microsoft AI, its founding leaders have left, and there will be no AIF internships next year. It's the backstory to this week's AutoGen drop to maintenance-only: the research org that built it no longer exists in that form.

  • AI APIs churned 60+ times in eight days

    One tracker counted roughly 35 price moves and 27 model delistings across providers in an eight-day stretch. Pin exact model snapshots and watch provider changelogs — an agent pinned to a floating alias can wake up to a 404 or a surprise bill.