Three Names, One Machine
Three companies handed you the same new colleague this autumn. Before you give one the keys to your inbox, read the paperwork beside the pitch.
Key · companies amber
You give it a job. It opens its own computer in the cloud, signs into the tools you already use, and comes back only when it needs you. If that feels familiar, you have watched the same demo three times in seven weeks.
Grok Bot arrived from xAI on the 11th of August 2026; Muse from Meta on the 8th of September; dots from OpenAI on the 29th. One product, three names.
I spent a morning with six documents: three launch pages and the documentation beside them. A paperwork read, not a field test; nothing of mine is connected.
The same machine, three names
All three sell one shape: an agent with its own cloud computer, your connected apps, a memory, a schedule, and a gate that opens on your approval.
None of them invented it. The pattern came from one open-source project, OpenClaw. Sam Altman hired its creator, Peter Steinberger, on the 15th of February 2026 to “drive the next generation of personal agents”. Five days later, Andrej Karpathy named the shape: “first there was chat, then there was code, now there is claw”.
The question is not who copied whom. It is what each wrote in the margins.
What the poster says. What the record says.
Each launch page makes the same promise in almost the same words. Then you open the documentation, and the promise acquires clauses. Read together, the clauses answer two questions: who else can read it, and what survives you leaving. xAI’s answer is other Bots; Meta’s is Meta; OpenAI’s is the data itself.
Interactive · the fine print
Same promise, three sets of paperwork.
xAI · Grok Bot
“Bots have their own computer.”
“All of your Bots share one cloud computer … Do not use separate Bots as a security boundary.”
Meta · Muse
“a dedicated, virtual machine (VM) that houses both the agent and a person’s data”
“It does not prevent Meta from accessing data when necessary to support, secure or operate the service.”
OpenAI · dots
“they have their own cloud computer”
“Disconnecting an app does not delete information your dot has already obtained from it.”
All quotes as published. Links in Sources.
Read alone, each clause is reasonable engineering. Together, the six documents are a price list for trust. xAI’s shared machine means a login for one Bot is a login for all. Meta’s defences are the deepest, and Meta can still reach your data until the Confidential VM ships. OpenAI’s record means what the dot learned outlives the app, and taking it back means deleting the whole dot.
Why it happened this way
The models converged, so the contest moved to everything around them. Each company ships the same machine using the distribution it already owns: Meta has the chats and WhatsApp; OpenAI has the desk, inside ChatGPT, Slack and Teams; xAI has the builders, through Cursor’s plans. What is left is trust, and trust is a paperwork problem.
Questions I would ask first
Before connecting anything of mine, three questions, out loud. Whose computer does it run on? Who else can read what it keeps? And where does the proof live when the work is done?
What would change my mind: a shared machine becoming a per-bot boundary, a deletion that deletes, the Confidential VM arriving as promised. The paperwork is dated, and dates move; when it ships, I will publish the delta.
The stars are the same; only the drawings differ.
When the next launch lands, one question matters: where does the proof live when the work is done?
Readout: the documents, and how to check them
- Confirmed from the primary documents, opened and quoted as published: xAI’s launch page of the 11th of August 2026 and its security documentation; Meta’s launch post of the 8th of September 2026 and its security write-up; OpenAI’s launch post of the 29th of September 2026 and its help centre page. Every quote in the interactive above is verbatim from these six.
- Confirmed: Peter Steinberger’s move to OpenAI, announced the 15th of February 2026 (Altman’s post; Reuters, TechCrunch and CNBC reports), and Andrej Karpathy’s “claw” line of the 20th of February 2026.
- Reported, not independently tested: that Muse shows its agent instruction files inside the app, and the pricing tiers. Neither is load-bearing here.
- Not confirmed: nothing in this piece rests on it. Where a line is a company’s promise about the future, such as the Confidential VM, it is labelled as a promise.
- Style note: dots is written lowercase by OpenAI; it is a name, not a typo.
- What would change the picture: the promised isolation and deletion semantics shipping as written. The six documents carry dates; this read is valid as of the 11th of October 2026.
Sources & references
- xAI, Introducing Grok Bot, 11th of August 2026.
- xAI, Approvals, security, and privacy, Grok Bot documentation.
- Meta, Introducing Muse, 8th of September 2026.
- Meta AI Research, How We Built Safety Into Muse, 8th of September 2026.
- OpenAI, Introducing dots, 29th of September 2026.
- OpenAI, Getting started with your dot, Help Centre.
- Reuters, OpenClaw founder joins OpenAI, 15th of February 2026.
- TechCrunch, OpenClaw creator Peter Steinberger joins OpenAI, 15th of February 2026.
- Andrej Karpathy, post on X, 20th of February 2026; and Simon Willison, Karpathy talks about claws, 21st of February 2026.
About to hand an agent your inbox? Ask for the paperwork first: email brandon@kreostudio.co.uk.
Reader signal
Was this useful?
Work with KREO Studio
AI engineering, data science and design architecture, from Plymouth to the wider UK.
Next Article
