Claude Now Edits Inside Google Docs, Sheets and Slides. Set It to Ask First.

Anthropic released Claude for Google Workspace on 6 October: a sidebar inside Google Docs, Sheets and Slides so you work with Claude on the open file instead of copying text into a chat window and back. It is a public beta for every paid Claude plan (Pro, Max, Team and Enterprise), installed from the Google Workspace Marketplace. An admin can roll it out to the whole domain or to chosen groups from the Google Admin console.

In Docs it edits sentences, restyles headings and proposes bigger rewrites as suggestion cards that highlight what would change. In Sheets it writes formulas, builds pivot tables and charts, and cleans data with Python (a programming language, run for you) and writes the result back. In Slides it makes new slides using the deck's own layouts and theme and checks for overlapping or unreadable text.

The setting that matters: "Ask before edits" shows an approval card for each change. "Accept all edits" applies them without pausing. It uses the same models, connectors and skills as your Claude account, and the Enterprise compliance controls apply to the add-on.

OpenAI started rolling GPT-6 out to ChatGPT Plus, Pro, Business and Enterprise on 7 October, with Free and Go accounts following from 8 October. Paid tiers get the GPT-6 Sol model, free tiers get GPT-6 Luna. Enterprise access depends on the workplace admin settings. The models behind Work and Codex do not change.

The visible change is called Intelligent UI. Instead of a wall of text, an answer can include charts, maps, forms, tappable buttons and small working tools built on the spot: a calculator, a bill splitter, a comparison laid out side by side. ChatGPT decides when a plain text answer is better.

It also starts answering while it is still thinking. OpenAI says answers that need a web search start 44% sooner than the previous model, on its own internal test.

Anthropic released Claude Haiku 5.5 on 7 October and calls it its fastest model. The price is $0.10 per million input tokens and $0.50 per million output tokens for prompts under 100,000 tokens, and five times that above. A token is roughly three quarters of a word, so reading a thousand-word email costs about a hundredth of a penny.

Anthropic pitches it for high-volume, cost-sensitive work: summaries, sorting and classifying, pulling data out of documents, live customer support and browser tasks, with named customers using it for CRM reporting, financial summaries and document question-answering at scale. Artificial Analysis, an independent tester, scores it 43 on its Intelligence Index and measures about 240 tokens a second. It is available to developers on Anthropic's platform, AWS, Google Cloud and Microsoft Azure.

Separately, on 8 October Anthropic published its 2026 usage policy, effective 12 November. The high-risk section now spells out that when Claude is used for recommendations affecting someone's health, finances, legal rights, livelihood or essential services, a qualified person must review the output and the affected person must be told AI was involved. Anthropic says the requirement is not new, only now stated explicitly.

Google opened its SynthID Detector to everyone on 7 October at synthid.com, in English worldwide. Upload an image, video or audio file and it checks for the invisible watermark Google's AI tools add, plus watermarks from OpenAI, Nvidia and Kakao, with Apple's to follow. It had been limited to media professionals.

Google says it has watermarked over 180 billion images and videos since 2023 and already handles about a million verification requests a day through Search, the Gemini app and Chrome.

The limit is in the design: it finds watermarks from those companies. A picture from another tool, a screenshot, or a heavily edited copy may show nothing, and "no watermark found" does not prove a real photograph. Treat a positive result as proof and a negative result as no information.