Weekly briefing

AI briefing, week 47/2026: Gemini in Gmail, Claude Meetings, Copilot

Reading time
12 minutes
Published
Topic guide
AI models
Written by
modernagents editorial teammodernagents is an AI automation agency for SMEs.
AI-generated image: Brass telegraph key with an indigo base plate on dark walnut

This week, AI features turn up in places where nobody ordered them. You open your inbox in the morning and a summary sits above a long customer email. In the video call with your client there is an extra participant called Claude, taking notes. Depending on your plan and settings, both are switched on out of the box, and in both cases it is better that you decide than that chance does.

Official sources in this article

  • European Union

An example: an engineering office in Leuven with 18 staff works with Google Workspace Business Standard, and four people also have Claude on the Team plan. Emma in the office answers the customer emails, and the project leads meet their clients in Google Meet. This issue covers news from 10 to 17 November 2026.

Key takeaways

  • Gemini in Gmail is gradually coming to Google Workspace users aged 18 and older. In the EU, Switzerland and the UK it only takes effect once smart features are switched on. Check where your domain stands.
  • Claude can take notes in Google Meet. On the Team plan this is switched on out of the box, including for calls with clients. Decide this as the owner before somebody tries it.
  • Microsoft rebuilt Copilot at its Ignite conference. Chat and Cowork now share one start page. For this, Microsoft describes a subscription per user and a consumption model for particularly hard tasks.
  • The EU published its guidelines on labelling AI content in the Official Journal. For certain older generative systems, the transition period for machine-readable marking and detection ends on 2 December 2026.

01 / 06

Gemini in Gmail: summaries and "Help me write" for all Workspace customers

Emma opens a complaint with fourteen replies, and above the thread sits a summary she did not ask for. Google is bringing two Gemini features in Gmail to all Google Workspace customers, for users aged 18 and older. The first one summarises long email threads at the top of the conversation, the second one is the non-personalised writing aid "Help me write". Until now, both were only available to consumer accounts and specific Workspace editions. The rollout started on 10 November, is gradual and can take up to 15 days (Google Workspace Updates).

Google says the features are on by default, and each user can switch them off. There is one condition. Smart features must be switched on in Gmail.

For businesses in Europe, this condition matters. For domains in the European Economic Area, Switzerland, the UK and Japan, Google says smart features are off by default. Your admin can set a default for everyone, and each employee can override the default in their own settings (Google Workspace Admin Help).

For the engineering office this means: if nobody changed anything, Emma does not see the summary. If she switched on smart features herself at some point, she can see the summary as soon as the rollout reaches her account, provided Gemini in Gmail is not switched off for her.

This is where it pays to act now:

  1. Ask your admin to check in the Admin console which default applies to smart features.
  2. Decide if Gemini may summarise customer emails, and write the decision into your AI usage policy.
  3. Tell your team that a summary does not replace the email. Before somebody answers a customer, they read the part of the original that matters.

02 / 06

Claude Meetings: Claude takes notes in Google Meet

Tuesday, 10 a.m., a site meeting with the client in Google Meet. The participant list has one more name, and a notice in the chat says that Claude joined the call. Anthropic is rolling out Claude Meetings as a beta on the Team and Enterprise plans. Claude joins a Google Meet call as a visible participant, takes notes, answers questions and builds documents during the conversation (Claude release notes).

For the owner, the defaults are what counts. On the Team plan, Claude Meetings is on by default and an owner can switch it off. On the Enterprise plan it is off until an owner switches it on. Claude currently works with Google Meet only, and Google Calendar must be connected. Each employee sets for themselves if Claude joins their own meetings without being asked.

Anthropic says Claude saves a transcript, a summary and descriptions of the shared screen, but no audio. The help article does not say how long this data is kept. On the Team plan, Claude may by default also join calls with outside guests, and an owner can switch this off. What the guests say then appears in the transcript, but they do not get the recap email (Claude Help: Use Claude Meetings).

Anthropic itself writes that everyone on the call should be okay with being transcribed before Claude takes notes. For the engineering office this is more than politeness. A site meeting contains names, prices and sometimes points of dispute, and the transcript holds personal data of the client.

If you have Claude on the Team plan, it pays to act now:

  1. As an owner, open the organisation settings and decide if Claude Meetings stays on, and for whom.
  2. Decide if Claude may join calls with clients. If you are not sure, switch off outside guests first.
  3. Agree a rule for internal meetings: whoever adds Claude says so at the start, and if somebody objects, Claude stays out.
  4. Check with the person responsible for data protection if transcripts of client calls fit your rules before you allow outside guests.

03 / 06

Microsoft Ignite: Copilot gets Home, and Office now lives inside Copilot

If your team works with Microsoft 365 Copilot, the Copilot app may soon look different. At the start of its Ignite conference, Microsoft made a new start page called Home generally available. The post does not say which users get it. Chat and Cowork sit side by side there, that is the conversation with Copilot and the area where Copilot works through longer tasks on its own. "Office in Copilot" comes on top and lets you create and edit Office files directly in the Copilot app (Microsoft Copilot Blog).

Microsoft writes that Office in Copilot is included in the Copilot licence. Which licence exactly is meant, and what users without a Microsoft 365 Copilot licence get from Home, is not stated in the posts from the first conference day. For businesses where nobody has a Copilot licence, the limits from week 46 still apply.

The money side deserves a second look. In Chat, Copilot now suggests on its own to hand a task to Cowork, and these recommendations are on by default. Files from Chat and Cowork are saved in OneDrive (Microsoft Tech Community). In the Ignite post, Microsoft describes a subscription per user and a consumption model that gives access to the most powerful AI for particularly hard tasks. For Cowork, Microsoft says your team sets a spending limit.

Much of the rest from the stage is not there for everyone yet. The personal agent Autopilot, the shared Workspaces and the daily overview "Today" start with customers in Microsoft's Frontier programme or in a private preview. A model selection called Auto is meant to pick the right model for each task in Cowork, also in Frontier first. Microsoft says employees can also choose manually between models from OpenAI and Anthropic, as far as the organisation's policies allow.

Planning is enough here. Before somebody on the team uses Cowork, ask your Microsoft partner which spending limit is set for you. The conference runs until 20 November, and we will look at prices and licence terms for smaller businesses next week.

Got a workflow like this?

Tell me about it. Honest take and a rough price, usually within a day.

AI models →

04 / 06

EU AI Act, Article 50: guidelines in the Official Journal, transition until 2 December 2026

A chatbot answers customer questions on your website, and the newsletter is written with AI. Do you have to label that? The European Commission published its guidelines on the transparency obligations in Article 50 of the AI Act in the Official Journal on 12 November. The guidelines are not binding, but they show how the Commission reads the rules. Article 50 has applied since 2 August 2026 (Official Journal C/2026/6005).

What is new in practice is a date. Article 50(2) generally requires providers of covered generative AI systems to mark the covered output in a machine-readable way and to make it detectable as AI-generated or manipulated, with exceptions. For such systems that were placed on the market or put into service before 2 August 2026, the guidelines describe a transition period for this until 2 December 2026. The amending regulation itself names in Article 111(4) only systems that were placed on the market before that date (Regulation (EU) 2026/1744). If you use a system that you developed yourself only in your own business, do not rely on the period without legal advice. This period does not apply to the duty in Article 50(1) to tell people in a direct conversation that they are talking to an AI. That duty has applied since August, with the exceptions the article names.

Most small businesses are not providers but deployers, because they use a finished system. The marking under Article 50(2) generally lies with the provider of the system. Deployers can have disclosure duties of their own. Anthropic shows this week what a marking looks like. Text from supported Claude models can carry an invisible watermark in the text itself on supported platforms, with exceptions, and Anthropic says it has signed the EU code of practice on labelling AI-generated content.

You should still know two cases. According to the Commission's example, a company that develops a chatbot in-house and puts it into service in the EU for its own use under its own name is itself a provider. And for published text that an AI generated or manipulated and that is meant to inform the public on matters of public interest, a disclosure duty generally applies, with exceptions. One exception requires that the text went through human review or editorial control and that a person or a company holds editorial responsibility. According to the guidelines, human review requires relevant knowledge and professional judgement, with fact-checking as a minimum, and a spell check alone is not enough.

We recommend planning here:

  1. Write down where in your business an AI talks to customers or creates text, images and video for the public.
  2. Ask the vendor of your chatbot in writing how the product handles the AI notice and the marking of its output.
  3. Record who reviews the substance of AI text before it is published.

What else the AI Act asks of small businesses is in EU AI Act for SMEs.

05 / 06

In brief: the other providers

Provider What is confirmed What you can do
OpenAI "Businesses in ChatGPT" started as a beta. Eligible businesses can claim their profile in Business Hub, correct details and add product data, subject to availability and verification. Before changes are published, an identity check through the service provider Persona is required. Business Agents start only for eligible ecommerce and online service businesses in the US with a website (OpenAI Help) Watch: check if your business is eligible if customers look for you in ChatGPT
OpenAI GPT-6.2 Astra and GPT-6.2 Sol have been in the API since 17 November (OpenAI changelog). At the standard rate and up to 272,000 input tokens, Astra costs USD 8 per million input tokens and USD 40 per million output tokens, Sol USD 2 and USD 10. Higher rates apply to longer contexts (OpenAI pricing) Watch: test on your own cases before a service provider switches
Google Gemini 4 Hydrogen has been generally available in the Gemini API since 12 November, Gemini 4 Carbon since 10 November. For Carbon, the introductory price of USD 2 per million input tokens and USD 10 per million output tokens applies until 30 November (Gemini API changelog) Plan: if your software runs on the Gemini API
Google Skills in Google Workspace reach Scheduled Release domains from 30 November. For users aged 18 and older they are on by default in the eligible editions, and for now they support only English instructions (Google Workspace Updates) Plan: decide beforehand who may create and share skills
European Commission The draft template for the fundamental rights impact assessment is open for feedback until 15 January 2027. The duty applies to certain deployers of high-risk AI and, according to the Commission, from 2 December 2027 (European Commission) Watch: only affects businesses that use high-risk AI

06 / 06

For your IT team

Super administrators set smart features in the Google Admin console under Account, Account settings and "Smart features for Google Workspace". There they set a default or leave the choice to the users. Changes can take up to 24 hours. Google says Gemini in Gmail can also be switched off for an organisational unit or group under Generative AI, Gemini for Workspace and "Feature access". Google's help page for this names only Enterprise Standard, Enterprise Plus and education plans under editions. So check if the switch exists in your edition (Google Workspace Admin Help).

Owners switch Claude Meetings under Organization settings, Products and "Claude Meetings (beta)". The options are off, on for specific roles or on for everyone. According to the help article, organisations with zero data retention, that is without lasting storage of inputs and outputs at Anthropic, do not get the feature.

According to the documentation, Anthropic's text watermark adds no tokens and no hidden characters and holds no information about your organisation (Anthropic documentation).

With Claude Opus 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 it can be missing if a request sets temperature, top_p or top_k to a value other than the default (Claude Platform release notes).

Anthropic says that anyone who builds Claude into their own product should assess for themselves what Article 50 requires of that product (Claude Help).

The Claude API has two more changes. With linked billing, a beta, customers of an app can link their own Claude organisation to the app if that organisation pays Anthropic directly for API usage. The developer then chooses who pays on requests to /v1/messages or /v1/chat/completions, and the customer can set a spend cap. Claude Managed Agents can now run as a whole service on your own infrastructure. The model requests still go to the chosen provider, by default to Anthropic.

If you move from Gemini 3.8 Flash to Gemini 4 Hydrogen (gemini-4-hydrogen), read Google's migration notes. The default for thinking_level is high, the value minimal is rejected, explicit caching is not supported, and thinking tokens are billed. For GPT-6.2, OpenAI requires the Responses API for tool calls.

On Amazon Bedrock AgentCore, the Web Search API is generally available. It costs USD 7 per 1,000 queries and runs in Europe in the Ireland region (AWS).

We also checked Mistral, Meta, Aleph Alpha, DeepSeek, Qwen, Cohere, xAI, n8n, Make and Zapier as well as the RTR KI-Servicestelle, the Bundesnetzagentur and the data protection authorities in Austria, Germany and Switzerland, and selected no item from them for this issue. Benchmark claims, stock market listings, funding news and cost studies commissioned by vendors stay out on purpose.

If you want to know what one of these changes means for your existing workflows, we are happy to talk it through in an AI consulting call.

Other issues of the AI briefing: week 41/2026 · week 42/2026 · week 43/2026 · week 44/2026 · week 45/2026 · week 46/2026.

ShareEmail

Related reading

AI models

Got a workflow like this?

Tell me about it. Honest take and a rough price, usually within a day.

Robert Van Ysendyck · Founder & Managing Director