Updated 21 min readBy Michael van Wijhe

AI news October 2026: what it means for your business

Blue Hello Growth cover with the text "AI 10/2026", below it Copilot, Claude, OpenAI and Gemini

In short

  • AI news from late September and October 2026 has one common thread: AI is moving from experiments to real work, and that calls for limits, permissions and an owner.
  • Twenty-two developments matter for businesses, from usage billing in Microsoft Copilot and GPT-6 in ChatGPT to the end of GPT-4o in the API, the first AI Act enforcement by the European AI Office and two new models from Europe's Mistral.
  • You do not need to act on all of it. If you use Microsoft 365, review your Copilot settings before 1 December. If your team uses ChatGPT, agree how you check charts and calculators it produces. If your team uses Gems in Gemini, make a list before they go away. If an integration runs on OpenAI's GPT-4o or GPT-4.1, switch before 14 January 2027.

November's AI news is in AI news November 2026.

There is AI news every week, but little of it changes anything for an organisation. From late September and October, twenty-two developments stand out because they affect what you pay, who does the work and how far you can trust an AI system.

The common thread: AI is shifting from an assistant that answers questions to agents that carry out work on their own. That saves time, but only if you cap the costs, decide what an agent may do and make someone accountable. Below, for each development: what happened, what it means and where to read more. This overview was updated on 9 October with GPT-6 in ChatGPT and Claude Haiku 5.5, on 11 October with three developments at Google, on 21 October with six developments from 5 to 14 October (items 10 to 15), and on 27 October with five developments from 14 to 26 October (items 16 to 20). That update also corrected item 8 on Gems and skills. Two developments from Mistral were added after that (items 21 and 22). Item 21 was updated on 30 October: the weights have been released and the price has been corrected. All of this month's Google news is also in Google AI news October 2026.

1. Microsoft Copilot: agents are billed by usage

On 25 September Microsoft announced a new Copilot, with Cowork for work you delegate, Code for building your own apps and Autopilot as an agent that keeps working in the background. It comes with a second pricing model. Chat and Copilot in Word, Excel, PowerPoint, Outlook and Teams stay under the fixed per-user licence. Agentic work is billed by usage, in Copilot Credits (Microsoft).

Microsoft does not publish a price per task, so you control costs through limits in the admin centre. One setting needs attention straight away: by default, new agents are added to your spending policy automatically. For new Business licences bought through a partner, usage billing will be on by default from January to March 2027, in all markets including the Netherlands and Germany. Microsoft announced this on 3 November; this item previously said it would not yet be on by default in those countries (Microsoft).

Read more: What will Microsoft Copilot cost now that agents are billed by usage? and Rolling out Microsoft Copilot in your team

2. Anthropic: the shortage is people

On 2 October Anthropic launched the Claude Frontier Academy, committing $100 million to train 10,000 engineers who can make AI work inside businesses. The company calls talent "one of the most pressing issues in AI implementation" (Anthropic).

That matches European figures. Among EU enterprises that considered AI but did not use it, lack of relevant expertise is the most common reason (Eurostat). The programme itself targets large enterprises, but the lesson applies to every organisation: decide how you will organise the expertise, and name an owner.

Read more: AI implementation: the bottleneck is people, not the model

3. OpenAI: reports of unwanted model behaviour

Since September, OpenAI has kept a public page of reports on cases where its models behaved differently than intended. The six reports currently listed include an agent that worked around internet restrictions and a model that published a researcher's credentials. According to OpenAI, all of them involved internal models and evaluations, not customers (OpenAI).

For an organisation this is no reason to panic, but it is a reminder. An agent that acts on its own can go further than you intended. Agree in advance what an agent may do by itself, what it must ask first and what it must never do. In late September OpenAI also paused the launch of a new model for similar reasons. What that means, and five questions to ask beforehand, are in Using AI agents safely.

4. Gemini Enterprise queries data without copying it

On 2 October Google released a preview that lets Gemini Enterprise query data directly in BigQuery, Cloud SQL, AlloyDB and Spanner. According to Google, this works "without copying or ingesting data into a data store", using each user's own credentials (Google Cloud). The connection runs over MCP (Model Context Protocol), an open standard that lets AI applications talk to other systems.

This is mainly relevant if your data already lives in Google Cloud. Two things matter. Data does not have to be stored again separately for AI, and an employee sees through AI only what they are allowed to see anyway. It is still a preview: test it, but do not build a business-critical process on it yet.

5. ChatGPT gets GPT-6, including the free tier

On 7 October, OpenAI brought GPT-6 to the ChatGPT chat screen on every plan. The new part is Intelligent UI: where it helps, ChatGPT answers with charts, buttons, forms and small calculators inside the conversation (OpenAI). Paid plans get GPT-6 Sol; Free and Go get the lighter GPT-6 Luna. On Enterprise, the admin decides whether the new interface is switched on.

For a team, this mainly calls for an agreement on checking. A chart or calculator looks finished, but it works with assumptions you do not always see straight away. And a tool you need every week does not belong in one employee's chat.

Read more: GPT-6 in ChatGPT: what Intelligent UI changes for your team

6. Claude Haiku 5.5: lots of small AI tasks get cheaper

On 7 October, Anthropic released Claude Haiku 5.5, its smallest and fastest model. It is built for large numbers of small tasks, such as summarising, sorting, querying data and customer support. It costs $0.10 per million input tokens and $0.50 per million output tokens for requests up to 100,000 tokens. Tokens are the chunks of text a model works in. Anthropic says this makes it around 75 per cent cheaper to run than Haiku 4.5 (Anthropic).

This mainly matters if you use AI inside your own software or processes, through the API. A task that costs a fraction of a cent each time becomes a budget line at thousands of runs a day. A cheaper model can make such a use case worthwhile. Test it on your own work first, though: the comparisons come from Anthropic itself. More on Anthropic's models in What is Claude? How to use it in your business. What the heavier Opus 5.5 from 22 September costs and when it is worth it is covered in Claude Opus 5.5: what is new, what it costs and what to change.

7. Google: the Gemini agent gets its own email address

On 8 October Google Cloud announced the Gemini agent, "a single, universal agent for work". You give it an objective rather than an instruction, and it works in Gmail, Docs, Sheets and Calendar, but also in Slack and Microsoft 365. A coworker agent gets its own Workspace account with an email address, calendar and Drive, acts under its own name and writes everything to a log. You set a hard spend cap per project. Gemini picks a model for each job, including Anthropic's Claude models (Google Cloud).

The agent is still in private preview, with no date for the wider rollout and no full pricing. Use the time to decide who owns an agent, what it may see and what it may cost.

Read more: Gemini agent: Google's AI colleague with its own email address

8. Google: Gems are becoming skills

Google is replacing Gems with skills: reusable instructions for recurring tasks. Skills have been arriving in the Gemini app since 13 October. The Workspace rollout was due to start on 5 October, but Google says it is delayed, with no new date (Google Workspace Updates). Google describes what happens to Gems on 17 November in two ways. According to Workspace Updates they move to the settings panel of the Gemini app, where you can still create and use them. According to Gemini Apps Help, Gems on personal accounts become read-only then and are removed on 17 December (Gemini Apps Help). On Workspace Business and Enterprise accounts you can no longer use Gems from no sooner than 1 March 2027. Whatever is left then becomes a draft skill that its creator has to activate (Gemini Apps Help).

Of everything in this overview, this is the only item with a hard deadline. Take stock of which Gems your team uses, and assign an owner for each Gem you keep.

Read more: Gemini Gems are becoming skills: what do you need to do?

9. Google: meeting notes without the cloud

On 6 October Google released AI Edge Foresight, a free, experimental Mac app that records meetings, fills out your notes and answers questions from your own documents. The AI runs on your own Mac, on the open Gemma 4 model and the new EmbeddingGemma 2 (Google Developers Blog).

For sensitive conversations that is appealing. Two caveats. Google does not say whether transcription handles Dutch, and according to heise the app does ask you to share data, while the AI works locally (heise). So test it first with a short conversation that contains nothing sensitive, and ask for consent before recording other people. If you want to take meeting notes with AI today, read AI meeting notes: three ways with ChatGPT.

10. OpenAI adds an invisible watermark to ChatGPT text in the EU

On 5 October OpenAI announced that text from ChatGPT and Codex will carry an invisible watermark for users in the EU over the coming weeks, on every plan. In the API it is off by default and you can opt in. The detector is not public: only researchers and expert organisations can apply for access (OpenAI). The reason is Article 50 of the AI Act, which requires providers to mark generated text in a machine-readable way.

Little changes for your team. You do not need to switch anything off, and a missing watermark does not prove a person wrote the text. Do record where you use AI and who approves a text.

Read more: Will AI text get a watermark? And can you detect ChatGPT text?

11. Meta: Muse Wonder, Meta's first frontier model

On 12 October Meta released Muse Wonder and Muse Wonder Full, its most capable models, built for agentic work. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full only to Muse Code subscribers. Meta puts the emphasis on memory: the model should retain and combine what you tell it over a longer period (Meta). According to TechCrunch, Muse Wonder costs $4 per million input tokens and $20 per million output tokens through the API, in the same bracket as the top models from OpenAI and Anthropic (TechCrunch).

For most organisations nothing changes yet. Meta does not say whether the models are available in the EU, and the comparisons with other models come from Meta itself. It does mean there is now a fourth provider of frontier models next to OpenAI, Anthropic and Google. If you test it, check first how Meta handles what you put in.

12. Dutch Data Protection Authority: a dedicated directorate for AI oversight

On 9 October the Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) announced a new directorate for AI and algorithm oversight. One department analyses the risks of high-impact algorithms; another prepares market surveillance under the AI Act and builds the expertise to test AI systems (AP). The AP has not yet been formally designated as market surveillance authority. The AP only gets those powers once the Dutch act implementing the AI Act has come into force. According to Binnenlands Bestuur, the European deadline for that designation already passed on 2 August 2025 (Binnenlands Bestuur).

So oversight is getting closer, even though the powers are not there yet. Part of the AI Act already applies, such as the prohibited practices and the duty to make sure staff have sufficient AI knowledge. Start there, because that is something you can arrange yourself.

Read more: AI literacy in your organisation and EU AI Act for SMEs: what you must and must not do

13. Google: Gemini 4 Argon and a tasks mode in the Gemini app

Since 13 October, Google AI Pro and Ultra subscribers can use Gemini 4 Argon in the Gemini app. New is the tasks mode: with a toggle, Gemini keeps working while you are away, in your Google apps and in more than forty other apps (Google). Argon is Google's new top model, announced on 30 September for long tasks with many steps (Google).

In Europe you mostly cannot use tasks mode: according to Google, tasks in Gemini are not available in the European Economic Area, Switzerland or the United Kingdom, nor with a work or school account (Gemini Apps Help). So it is not yet an option for a business process. To work out when an agent like this does fit, read What is an AI agent?

14. OpenAI: European SMEs use agents eight times as much

On 14 October OpenAI published figures on the use of its agentic tools in the EU. Among SME customers (up to 200 users), weekly users of Codex and ChatGPT Work have grown more than eightfold since May, with growth in every member state. At the same time, only 10 per cent of the leading firms worldwide are in the EU (OpenAI).

Read these as a vendor's figures about its own paying customers, not about all businesses. The lesson is still useful: the gap is not access to AI, but how firmly AI is built into the work.

Read more: AI agents in SMEs: is Europe falling behind?

15. OpenAI retires GPT-4o and GPT-4.1 in the API

On 14 October OpenAI announced that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from the API on 14 January 2027. The recommended replacement is GPT-6 Luna. Fine-tuned versions of these models stop on the same date (OpenAI).

If you only use ChatGPT in the app, you will not notice. If an integration, automation or chatbot runs on one of these models, it will stop unless you switch. Find out now which models you use and test the replacement on your own examples.

Read more: Will your AI integrations stop on 14 January? What to do now

16. OpenAI: GPT-6.1 Terra works in screens, at a lower price

On 26 October OpenAI released GPT-6.1 Terra, the middle model in its line-up. It is built to understand what is on a screen, take actions across apps and write naturally. In ChatGPT Work and Codex it is rolling out to Plus, Pro, Enterprise and Edu; the announcement does not mention Business. Through the API it costs $1 per million input tokens and $5 per million output tokens (OpenAI). OpenAI says it matches GPT-6 Astra at a tenth of the price per token and finishes tasks on average three times faster than Astra and GPT-6.1 Sol. For complex coding and in-depth research, OpenAI still recommends Sol or Astra (OpenAI).

This is mostly of interest for automations and agents that work in screens and forms, where speed and cost per task matter. The comparisons come from OpenAI itself. If you have to move off GPT-4o anyway (item 15), test Terra alongside the recommended replacement, GPT-6 Luna, on your own examples. More on the other models of this generation in GPT-6.1 Sol, Ultrafast and the Agents API: what to choose.

17. Anthropic: Claude Duet brings work to you

On 21 October Anthropic opened a research preview of Claude Duet, an app in which Claude works as a proactive agent. Duet looks through your connected apps, sends a short morning brief of what needs attention and checks those apps every few hours. It carries out tasks itself, but before it sends a message, makes a payment, signs or deletes anything, it shows exactly what it is about to do and waits for your approval. Some actions, such as moving money between accounts, are off-limits even with approval. Duet runs on Claude Opus 5.5 and is available through a waitlist, for now only to Pro and Max subscribers in the United States (Anthropic).

So you cannot use it in Europe yet. It does show where the market is heading: after OpenAI with dots and Google with the Gemini agent (item 7), Anthropic now also has an agent that takes the initiative. The design revolves around the moments when the agent has to stop and ask. That is exactly what you need to set down yourself before you let such an agent into your work.

18. Agent Skills becomes an open standard at the Linux Foundation

On 22 October Anthropic donated Agent Skills to the Agentic AI Foundation, part of the Linux Foundation. Agent Skills is an open format for skills: written instructions and ways of working that an AI agent uses for a recurring task. Anthropic says more than seventy platforms already support it, including ChatGPT, Cursor, GitHub Copilot and VS Code. Engineers from Anthropic, Google, OpenAI and others maintain the standard (Anthropic).

For an organisation, this means a well-described way of working is less tied to a single tool. Anthropic does not say whether skills in Gemini (item 8) follow the same format. The advice therefore stays the same: keep recurring instructions in a document you control, with an owner, and load them from there into whichever tool your team uses.

19. European Commission: the AI Office starts enforcing

In mid-October, Henna Virkkunen, Executive Vice-President of the European Commission, told the International AI Summit in Dublin that the Commission has taken its first enforcement actions under the AI Act. She referred to incidents in which AI agents escaped their environment, inserted malicious code and deceived people. The Commission's AI Office sent requests for information on issues including security, external evaluation and monitoring how models are used (European Commission). On 9 October the Commission had already held a special meeting of the AI Act's scientific panel on these incidents (European Commission). The Commission does not say which companies received the requests.

This concerns the makers of the most powerful models, not businesses that use AI. A request for information is not a fine either. What applies to you stays the same: make sure your team has sufficient AI knowledge, avoid prohibited practices and be open about AI content. Two more prohibitions apply from 2 December 2026. What that means for a small business is in EU AI Act for SMEs.

20. Google: Gemini conversations in Vault and meeting recaps in Chat

On 22 October Google added Gemini conversations to Google Vault, the archiving tool in Workspace. Admins can filter and export the conversations staff have with Gemini in Gmail, Drive, Docs, Sheets, Slides and Meet, for example for a legal investigation. This only covers conversations created while conversation history was on. Retention rules and litigation holds for these conversations will follow later (Google Workspace Updates). In addition, since 16 October Google Meet posts a recap with decisions and action items in the meeting's Chat space after a meeting, if "Take notes for me" was on (Google Workspace Updates).

If you manage Workspace, check whether Gemini conversation history is on for the teams that need archiving. The Chat recap is on by default. Make sure your team knows that decisions and action items become visible to everyone in that space. If you meet in Dutch, it will not help much: Dutch is not on the language list for "Take notes for me" (Google Meet Help). Alternatives are covered in AI tools for your business and in item 22.

Read more: Google AI news October 2026

21. Mistral Large 4: a European frontier model

Mistral, the French AI provider, released a preview of Mistral Large 4, its largest model, on 6 October. It handles text and images, supports more than 160 languages and is built for reasoning and agentic work. Through the API it costs $0.70 per million input tokens and $2.10 per million output tokens (Mistral). On 29 October Mistral released the weights under a modified MIT licence. That lets you run the model on your own servers, commercially too; only if you sell the model itself as a service do you need a separate licence (Mistral).

For organisations that want to know where their data is processed, this is the most important European development of the month. The comparisons with other models come from Mistral itself. Test the API on your own work first. When a European model fits and which route to take is covered in Mistral AI: when should you choose a European AI model?.

22. Mistral Voxtral 3: speech to text in Dutch

On 20 October Mistral released Voxtral 3, a family of speech models. They turn speech into text, including Dutch, recognise who is speaking and can handle specialist terms you supply in advance. There is also a text-to-speech model. You use Voxtral through an API hosted in the EU, or on your own servers. Transcribing a recording costs from $0.0035 per minute (Mistral).

It is not an app for end users but a building block. That makes it interesting alongside the meeting tools: Google Meet's "Take notes for me" does not support Dutch (item 20). If you want customer calls, intakes or meetings in Dutch transcribed within the EU, an integration with a model like this can be an alternative. Test the quality on your own recordings first, and ask for consent from the people you record. Which routes keep your data in the EU is covered in Transcribing Dutch with AI.

What does this AI news mean for your business?

DevelopmentFor whomWhat to do now
Copilot usage billingLeadership, IT, financeSet limits and a pilot group before you start
AI talent shortageLeadershipChoose an owner and a route per process
Unwanted model behaviourLeadership, ITDefine what agents may do on their own
Gemini EnterpriseData and ITTest only if your data is in Google Cloud
GPT-6 in ChatGPTEveryone using ChatGPTAgree how to check charts and calculators
Claude Haiku 5.5IT and developersTest for high-volume small tasks via the API
Gemini agentLeadership, ITDecide on owner, permissions and spend cap before the rollout
Gems become skillsEveryone using GemsTake stock of Gems and choose an owner for each
AI Edge ForesightAnyone meeting on a MacTest with a short Dutch conversation
Watermark in ChatGPT textAnyone publishing AI textRecord who approves AI text
Meta Muse WonderIT and developersWait until EU availability is clear
AI oversight at the APLeadership, complianceSort out your team's AI literacy
Gemini tasks modeNobody in the EEA, Switzerland or the UKWatch, not yet usable
Agents in European SMEsLeadershipPick one process where AI becomes part of the work
GPT-4o retired in the APIIT, integration ownersSwitch and test before 14 January 2027
GPT-6.1 TerraIT and developersTest for agents that work in screens
Claude DuetNo one in Europe: US onlyDecide where an agent must ask first
Agent Skills as an open standardAnyone using skills or fixed instructionsKeep instructions in a document you own
Enforcement by the AI OfficeManagement, complianceOwn duties in order: AI knowledge, prohibitions, transparency
Gemini in Vault, recaps in ChatWorkspace adminsReview conversation history and Chat recaps
Mistral Large 4IT, anyone keeping data in EuropeTest the API on your own tasks, then choose per task
Mistral Voxtral 3Anyone transcribing Dutch conversationsTest on your own recordings, ask for consent

When can you ignore this?

If you mainly use AI personally, for writing and analysis, and have no agents or integrations running, little changes for you right now, apart from ChatGPT answering more often with charts and calculators. If you have had software or automations built for you, do check whether they run on GPT-4o or GPT-4.1. The news becomes relevant once you start handing work to AI. That is also the moment to sort out costs, permissions and ownership.

What is a good next step?

Pick one process with a lot of manual work and look at which of these developments changes anything there. Not sure which tool fits which job? Read AI tools for your business: which one for which job?

Sources

Newsletter

The AI news that matters, once a month

One email a month with the most important developments in AI and what they mean for your organisation. No noise.

You will receive the Hello Growth newsletter, about once a month. First you get an email to confirm your subscription. You can unsubscribe at any time with the link under every newsletter. Read how we handle your data.

Let us begin

Want this in your own organisation?

We build it, or we teach your team to. Either works.