It has been an unusually busy stretch for Anthropic. Within the span of a few days, the company flipped on a global watermarking system for Claude output, squeezed its Cowork agent into the Chrome sidebar, and watched a Chinese rival publicly assemble a team to chase its flagship coding product. Here is what happened, story by story.
Every Claude Response Now Carries an Invisible Mark
Anthropic announced on August 11 that Claude-generated text and images are now watermarked worldwide, with no opt-out. The move is part of the company's commitments under the EU AI Act's Article 50 Code of Practice on Transparency of AI-Generated Content, a pledge signed by nearly 200 companies including Meta, Microsoft and OpenAI.
The mark is applied at the model level: an imperceptible, machine-readable pattern is woven into generated text. It travels with the content when copied and pasted, and it may survive light editing. Image files such as PNG, JPG and SVG receive digitally signed provenance metadata based on the C2PA open standard, so anyone can check whether a file passed through Claude and whether it has been tampered with.
Because the marking happens inside the model, it shows up no matter which product produced the text:
- Claude chat and Claude Code
- Claude Cowork and API access
- Claude served through AWS, Google Cloud or Microsoft Foundry
Models released on or after August 2 carry the watermark from launch; support for older models is still rolling out. Anthropic is upfront about the limits. A detected mark is a signal, not proof: people use Claude to proofread, translate or summarize their own writing, and that output carries a Claude mark even when the ideas are entirely human. The absence of a mark proves nothing either, since heavy editing, paraphrasing, short passages and format conversion can all strip it. There is no official removal tool, and detection tools for users and third parties are still in development.
That nuance has not calmed the reaction. Users on X and Reddit called the policy "hugely problematic" and declared the end of "the era of unprovable AI writing." TechCrunch reports that many Claude users are angry the watermarks will catch them using the assistant at jobs and classes. Coders worry that cryptographic signatures will degrade generated code, while writers who use Claude as a proofreader feel the most betrayed. Radio host Erick Erickson summarized the grievance: "I had ditched Grammarly for Claude for proofreading because it does a better job. But now the stuff I've written will be watermarked that Claude did the work. This is ridiculous."
The policy exists because of Europe. Google has watermarked AI images since 2023 and has since expanded to text, audio and video. OpenAI, per the Wall Street Journal, has had text-watermarking technology for years and has chosen not to ship it. Anthropic is now the biggest name to turn it on for everything, everywhere.
Cowork Moves Into Chrome's Sidebar
On the product side, Anthropic shipped a genuinely useful update: Claude Cowork can now run inside a Chrome sidebar. The feature lives in the Anthropic browser extension, giving users a full Cowork session without leaving the tab they are working in.
Cowork is Anthropic's agent for everyday computer tasks, distinct from Claude Code, and it expanded to mobile and the web in July. The Chrome integration is the latest step in making the agent a permanent, always-available layer across a user's digital workspace. For anyone juggling a document in the browser and a task list in a chat window, the sidebar version removes one more excuse to tab-switch.
DeepSeek Assembles a Claude Code Challenger
Meanwhile, the competition to dethrone Claude's coding agent just got louder. Bloomberg reports that DeepSeek has set up an official account for a new "DeepSeek Harness Team" on WeChat, verified by Tencent and registered to a Beijing entity that Chinese corporate database Qichacha identifies as controlled by DeepSeek.
The term "harness" refers to the software scaffolding wrapped around a large language model that lets an agent manage complex, multi-step work, which is exactly the layer that made Claude Code famous. DeepSeek is recruiting for the team now, and one job posting says the goal is to "turn its models into cutting-edge agentic products." The startup also updated its flagship V4 Pro model on August 11 with better agentic capabilities.
Early moves from the harness team include:
- A public WeChat presence for recruiting and announcements
- Dedicated hiring for agent-scaffolding engineering roles
- Agentic upgrades to V4 Pro, its top commercial model
Chinese firms are locked in a heated rivalry with US developers over building more advanced, more affordable AI tools, and the emerging market for business agents is the current battleground. DeepSeek has already undercut Western pricing on models; now it is publicly signaling that Claude Code's territory is next.
One more note from a busy week: a Claude outage on August 12 briefly affected multiple models, flagged by Anthropic's status page at 13:50 UTC and resolved by 18:07 UTC. For the moment, the bigger story is transparency. Watermarks change what the world can prove about AI-written text, and rivals are racing to match the agent experience hiding behind it. Claude's week was a preview of both forces at once.
Comments