Weekly Recap · Week 45/2026

AI Recap: The week Microsoft's Copilot limits and Google's video price hit small and mid-sized businesses

A week in which Microsoft caps Copilot Chat for users without an add-on license, Microsoft and Anthropic want to bring Claude closer to the EU Data Boundary, and Google lowers the price of its video model. There is also an EU debate that would base AI training on legitimate interest, and a security model from OpenAI that for now is only available to closed circles. For you, this means: check your licenses, clarify where your data is processed, and keep an eye on prices.

The week in one sentence

Microsoft is tightening the limits on Copilot Chat, Anthropic is adding cost centers and admin controls for Enterprise, and Google is making video AI 30 percent cheaper.

Copilot Chat gets usage limits, and Microsoft does not disclose the numbers

Microsoft is limiting how much Copilot Chat users without a Microsoft 365 Copilot license can use. For organizations with up to 2,000 Microsoft 365 seats, the limit has been in effect since November 2. For larger organizations, the start is scheduled for November 9. Microsoft does not publish the specific caps, and admins can neither view nor change them.

For you, this means: if your team uses Copilot Chat without a license, they can hit a limit in the middle of the workday. This week, check which teams actually need Copilot before license requests start landing in your inbox. Plan for the licenses in your budget instead of hoping the free feature will last.

Source: Microsoft Partner Center: November 2026 announcements Source: Microsoft Message Center: MC1536163, Limits in Copilot Chat (mirror at mc.merill.net)

Claude is set to run in the EU Data Boundary in Copilot by year-end

On November 5, Microsoft announced that Claude models in Copilot, including Copilot Cowork, will be processed within the EU Data Boundary for organizations in the EU and EFTA by the end of 2026. The data already resides there, but until now Claude processing took place outside the boundary. In the EU, the option is disabled by default, and the feature is not yet live.

For you, this means: the argument that Claude fails on data residency will soon lose its weight. You should still only enable it after a review, with an admin switch, a data processing agreement, and an internal rule on which data may go into which model. Set a date for this now, before the year-end deadline forces the decision.

Source: Microsoft Copilot Blog: Bringing Claude models in Microsoft Copilot into the EU Data Boundary, November 5, 2026

The EU Council wants to base AI training on legitimate interest, but it is not law yet

On October 21, the EU member states agreed on a negotiating mandate for the Digital Omnibus. Under it, training AI on personal data could be based on legitimate interest, without prior consent. In addition, the duty to report data breaches would only apply to cases likely to result in a high risk, and the reporting deadline would rise from 72 to 96 hours. A position from Parliament is expected in early 2027, and until then none of this applies.

For you, this means: if you put customer data into AI tools, you still need consent or a well-documented legal basis. The Council also emphasizes that AI development alone does not establish legitimate interest, so case-by-case review remains mandatory. If you need an overview of the AI Act, you will find it in the EU AI Act guide for SMBs.

96 hours Deadline for reporting data protection breaches under the Council mandate, up from 72 hours, and reportable only when there is a high risk.Borncity, November 8, 2026 · Council mandate of October 21, 2026
Source: Borncity: Digital Omnibus, EU Council wants to base AI training on legitimate interest

Anthropic delivers Enterprise controls and cost centers for Claude

On November 5, Anthropic released new security and admin features for Claude Enterprise, including a Security Center and central management, some of them still in beta. On November 6, the beta for project tags followed, which let Enterprise customers assign spending to a cost center, billing code, or matter number.

For a law firm, an agency, or a service provider, this means you can see which client causes which AI costs, which simplifies billing and controlling. However, the benefit depends on whether you already use Enterprise plans. If you only have individual accounts, you can wait with the switch for now.

Source: Anthropic: Claude Release Notes, November 2026

Google makes Gemini Omni 1.2 Flash generally available and cuts the video price by 30 percent

On November 5, Google made the video model Gemini Omni 1.2 Flash generally available through the Gemini API and Gemini for business. Clips are now 15 seconds long instead of 10, and you can add changes through chat. The price is about $0.07 per second for 720p video, 30 percent below the predecessor Omni 1.1.

For you, this means: product videos and social clips get cheaper, but only if the workflow is in place. A 15-second clip costs about $1.05 on paper, before any revisions. Test the model on a real campaign clip before you shift agency budgets.

$0.07 List price per second for 720p video in the Gemini API, after a 30 percent reduction.Google, November 5, 2026 · Announcement of Gemini Omni 1.2 Flash
Source: Google: Gemini Omni 1.2 Flash, 15-second video and lower pricing

OpenAI launches GPT-6 Cyber, but only for approved security teams

On November 5, OpenAI introduced GPT-6 Cyber, a model for vulnerability research and exploit development. Access runs through the Daybreak Red program, and you also need approval for this specific model. OpenAI describes the model mainly as better at finding and fixing vulnerabilities, and as more reliable at staying within the boundaries set by researchers. On the ExploitGym benchmark, it reaches 48.5 percent, while its predecessor GPT-5.6 Cyber scores 32.1 percent.

For small and mid-sized businesses, this is more of a signal than a tool. You cannot get the model with a simple API key. Approved security service providers, however, get a stronger tool for testing. If you run AI agents with access to your own systems, you will find the basic rules in the security guide for SMBs.

$20 List price per million input tokens for GPT-6 Cyber. Output costs $100.OpenAI, As of November 6, 2026 · API pricing page
Source: OpenAI: GPT-6 Cyber, announcement Source: BleepingComputer: OpenAI's new GPT-6 Cyber AI finds 81 Firefox flaws, 948 Linux bugs

What stays with you from this week's business

First: licenses are no longer a side issue. From November 9, the Copilot Chat limit is also set to affect larger companies, and you should know who uses Copilot without a license.

Second: data location is becoming a selling point. Claude in Copilot is set to run in the EU Data Boundary by year-end, but it is not live yet, so plan with a buffer instead of relying on promises.

Third: prices are moving in both directions. Google charges $2 per million input tokens for Gemini 4 Argon until November 30 and $4 afterward, so test now and put the deadline in your calendar.


Sources & further reading

As of November 9, 2026. Information based on statements from the companies and the cited media, without guarantee. Prices are list prices in US dollars according to provider websites. No reliably dated news on Mistral or Meta was found this week, so neither is included.

Frequently asked questions about the AI week

Does the Copilot Chat limit also apply to us?
If your employees use Copilot Chat without a Microsoft 365 Copilot license, yes. For organizations with up to 2,000 seats, the limit has applied since November 2. For larger companies, November 9 is the announced start. Microsoft does not publish the limits, and admins cannot adjust them.
Can I already use Claude in Copilot within the EU Data Boundary?
Not reliably yet. On November 5, Microsoft announced that processing will run within the EU Data Boundary by the end of 2026. Until then, the option is disabled by default in the EU, and according to reports, the feature is not yet live.
What does GPT-6 Cyber cost, and can I use it easily?
According to OpenAI's pricing page, the model costs $20 per million input tokens and $100 per million output tokens. Access, however, is tied to the Daybreak Red program and an additional approval. A regular API key is not enough.
What does this AI week mean for small and medium-sized businesses?
This week shifts costs and responsibility a step further toward you: Copilot licenses become a budget question, and data location and consent become a duty to check. If you sort out licenses, data flows, and model approvals now, you will not have to scramble later in the quarter.

How many of your employees have already hit the Copilot Chat limit?

In the free 60-minute workshop, we will look together at your Copilot, Claude, and automation approvals and sort out what needs to be done right away. No preparation required, and no slide marathon.

Request a workshop