OpenAI's always-on agents are blocked for European Pro users and shipped to European Business Premium users — at roughly the same price

A different kind of European launch gap

The familiar version goes like this. A vendor ships an AI feature in the United States, the press release says “more regions soon”, European users discover a greyed-out menu item, and someone at the company tells a reporter that the regulatory environment is complex. Gemini 3.8 Flash’s TTS voice replication went that way. So, in a more interesting fashion, did GPT-6 Astra, where EU data residency and the fast service tier turned out to be mutually exclusive.

dots is not that. The exclusion is narrower and stranger, and the strangeness is the story.

As reported from OpenAI’s release notes, Pro access to dots “excludes the European Economic Area, Switzerland, and the UK at launch.” In the same notes, Business Premium subscribers are described as getting dots “across all supported ChatGPT regions” — and ChatGPT Business is supported throughout the EEA. Enterprise, Education and Healthcare workspaces receive a beta that stays disabled until an administrator switches it on.

So consider two people in Berlin. One pays for ChatGPT Pro. The other is on a Business Premium seat at a two-person company. Same city, same GDPR, same AI Act, same DSA designation over the same product. One of them cannot have an always-on agent at any price this week. The other turned one on Tuesday.

The price does not explain it either

The easy assumption is that the business route is the expensive route, and that Europe is simply being asked to pay enterprise rates for a consumer feature. The numbers do not support that.

A European buyer paying $500 a month for ChatGPT Pro 500 — the tier whose pitch includes an included dot — is excluded. A European buyer paying $100 a seat on an annual Business Premium commitment is not. The exclusion gets more absolute as the consumer price goes up, which rules out cost recovery as the explanation and rules out capability too. The agent is the same agent. It runs on the same GPT-6 Astra, in the same cloud, on the same infrastructure.

One frictional detail matters for anyone planning to take the business route: a ChatGPT Business workspace requires at least two seats, and the plan is scoped to teams of 2 to 200. A solo European developer cannot simply convert. Two can.

What is actually being allocated

OpenAI has not cited a regulation for the dots exclusion, which is to its credit — vague gestures at Brussels are the standard move and would have been easy here.

Reading the facts rather than the absence of an explanation: the difference between a European Pro subscriber and a European Business Premium subscriber is not jurisdiction. It is that the second one comes with a data-processing agreement, a named controller, administrative controls, an audit trail, and a counterparty that is itself accountable for how the tool is used inside its organisation.

Now consider what a dot is. An agent that runs continuously on its own cloud computer, reads across every app a user has connected, and — in OpenAI’s own framing — reviews that information proactively and forms memories from it even when the user has not asked a question about it. Stack that against European obligations and the list gets long fast: a lawful basis for processing that keeps working when nobody initiated the processing, purpose limitation over data the agent decided on its own to read, data-minimisation applied to a memory the user never saw being written, and the Article 50 transparency duties sitting on top. ChatGPT also carries a Very Large Online Search Engine designation under the DSA since 31 August, with systemic-risk assessment duties attached.

In a business workspace, a large share of that burden has a home: the customer is the controller, the DPA governs the processing, and the admin who flipped the switch made a documented decision. In a consumer subscription, every one of those questions lands on OpenAI.

That is a defensible engineering and legal sequencing decision, not an outrage. What makes it worth writing down is the consequence, which is that capability in Europe is now a function of contract type. The same pattern already governs data residency, consent-or-pay advertising in the EEA, and service tiers. It has now reached the agent layer, which is where the next three years of product differentiation is going to happen.

The containment envelope deserves credit

A fair reading has to note what OpenAI did not ship.

Proactive research is restricted to read-only tools: a dot working on its own cannot send messages, cannot modify content in connected apps, and cannot control devices. Access to a local computer is optional and off by default. Texting is a limited beta for US Pro users only, routed through a third-party provider.

That is the conservative configuration, and it is the right first release for a product whose entire premise is operating without supervision. Compare it with Meta’s Muse, which shipped with an elaborate per-user sandbox that Meta’s own published Agents Rule of Two advises against in exactly that configuration — the contrast is instructive. Read-only is a weaker product and a much better starting posture, and the limits are published rather than discovered.

The open question is what happens when the read-only constraint comes off, because a dot that can send mail and modify documents is the version people are actually paying for. Whether that version arrives in Europe on a consumer plan is the thing to watch, and nothing in this launch suggests the answer is yes.

A continent without agents, for now

The pattern across vendors in 2026 is hard to miss.

• Meta Muse — launched in the US in September on iOS, Android and the web. No announced European expansion.
• Apple Siri AI — available on Mac and Vision Pro in the EU, absent from iPhone, iPad and Apple Watch. Apple has pointed at the Digital Markets Act; a Commission spokesperson responded that there is “absolutely nothing in the DMA” stopping Apple from launching new products in the EU.
• OpenAI dots — EEA, Switzerland and the UK excluded for Pro, available on Business Premium, no dated consumer rollout.

Three vendors, three different stated postures, one outcome for a European consumer. And in OpenAI’s case specifically, the Business Premium carve-out undercuts the strongest version of the regulatory argument: a rule that genuinely prohibited the processing would not have a $100-a-seat exception.

What Europe has is not a prohibition. It is a market where the agent layer is sold to accountable organisations first and to individuals later, and where “later” is undated.

What a European team should do this week

If you want dots and you are on Pro, the route is a ChatGPT Business workspace with Premium seats, not a waiting list and not a VPN — account region is tied to the plan and the workspace, and circumventing it puts the subscription at risk for no capability gain. Verify availability on your own workspace’s plan page before committing to annual billing, because rollout wording changes.

If you already run a Business or Enterprise workspace, dots is a decision you now own rather than one OpenAI made for you. The Enterprise beta is off until an admin enables it, which is the correct default and also means somebody has to assess it. Do that assessment against the read-only envelope described above, document which connected apps are in scope, and remember that memory formation is the part that happens when nobody is watching. The computer-use agent lineage is the closest existing analogue in your risk register.

If you are choosing a stack now, weight contract type more heavily than you used to. The agent tool roundup and the ChatGPT breakdown track what each plan currently includes, and the Claude comparison is the right place to check whether a rival’s agent layer has the same consumer-versus-business split. For European buyers the question to ask a vendor is no longer “is this available in my country” — it is “which of your plans is it available on in my country, and what changes if I move up.”

The finding

OpenAI shipped a genuinely new product category this week and gated it carefully, which is the responsible version of a risky launch.

It also drew the gate in a place that tells you something about the next few years. The line between a European user who can run an always-on agent and one who cannot is not drawn at the border, not at the price, and not at the capability. It is drawn at the signature.

Questions or corrections? Email Pick Right. Want the full list? See all news.