Personal AI in depth
Muse and Dots
Meta Muse and OpenAI Dots are competing to be your always-on personal agent. We cover pricing, safety design, integrations, and the surveillance and privacy trade-offs.Oct 3, 2026
In autumn 2026 the battleground shifted from "chat" to always-on personal AI agents. Meta's Muse and OpenAI's Dots lead the pack. Both research, act and sometimes buy on your behalf. Here is what is verified — plus the pricing, privacy and surveillance questions underneath.
- Verified Confirmed in official announcements or primary documents
- Independent check Confirmed in reporting or hands-on reviews
- AIfrontier view Our interpretation
1. What is a personal AI? Not a chatbot
A chatbot waits until you ask. A personal agent inverts that: give it a goal once and it keeps working — opening a browser, filling forms, negotiating, and checking back for approval on sensitive steps. WIRED put it bluntly: "The chatbot is dead, at least as a hot form factor for developers." Source: WIRED
2. Meta Muse: guarded by its own VMVerified
On September 8, 2026, Meta introduced Muse, powered by its Muse Spark model. Its defining choice is to house the agent and the user's data together in a dedicated virtual machine, Muse Secure VM. Source: Meta
| Item | Detail |
|---|---|
| Safety design | Runs on Muse Secure VM; a separate Sentinel agent is isolated at the system level so nothing Muse does reaches the internet without approval |
| Credentials | Muse has no visibility into passwords or payment methods; it uses them from secure storage |
| Approvals | Checks before sensitive actions (sending email, purchases) and shows a full audit trail |
| Payments | Checkout via Stripe's Link; the first AI agent covered by Link's purchase protections; one-time-use cards hide real card details |
| Training | Users can opt out of their interactions training Meta's models; data is not shared with ad systems |
| Pricing | Free for most needs; subscriptions at $20 or $100 a month by usage |
| Availability | US, on iOS, Android and the web; later on Ray-Ban Meta glasses. A Confidential VM, where only the user holds the key, is planned |
Independent check Muse topped US app download charts within two weeks, and Meta is testing a human concierge, Reuters reported. Source: Reuters
3. OpenAI Dots: always-on, and enterpriseVerified
At DevDay 2026 (Sep 29–30) OpenAI announced Dots, always-on personal agents powered by GPT-6 Astra. You message them through ChatGPT, Slack and Microsoft Teams, and Pro users can join a waitlist for iMessage / RCS. Source: WIRED
| Item | Detail |
|---|---|
| Pricing | Rolls out to ChatGPT Pro ($100/month). "We're starting out as a premium product. It uses a lot of compute," Altman said |
| Behavior | Crawls the web, completes tasks autonomously, pulls context from connected apps and learns preferences |
| Approvals | Explicit approval before sensitive actions like installing software or changing a password; a Custom Rules tool sets boundaries |
| Enterprise | "Specialist Dots" tuned for accounting, email marketing and legal analysis |
| Multiplicity | One Dot today; multiple Dots expected later |
4. ComparisonIndependent check
| Item | Meta Muse | OpenAI Dots |
|---|---|---|
| Announced | Sep 8, 2026 | Sep 29, 2026 (DevDay) |
| Foundation model | Muse Spark | GPT-6 Astra |
| Pricing | Free + $20/$100 a month | ChatGPT Pro $100/month |
| Reach | App, WhatsApp, later glasses | ChatGPT, Slack, Teams, iMessage/RCS |
| Safety core | Dedicated VM + Sentinel, payment protection | Approvals + Custom Rules, enterprise tuning |
| Enterprise | — | Specialist Dots |
5. The trade-offsAIfrontier view
| Issue | Detail |
|---|---|
| Privacy | Agents ask to connect email and Drive. WIRED's tester wants the tool to "earn more of my trust" before automating life |
| Surveillance | Muse pinged one user about a utility bill that had never been a problem — the helpful-vs-incessant balance is unsettled |
| Price | Dots starts at $100/month; Muse's free tier is its edge, with paid tiers by usage |
| Safety | With agents going on rogue hacking sprees elsewhere, approvals and boundaries are prerequisites |
| Lock-in | Personal data becomes the differentiator; with everyone entering at once, some warn of a painful fight over users |
6. Outlook
Autumn 2026's personal AI is a phase of lowering defenses with cute design while embedding always-on into daily life. What decides the winners is less raw capability than how far you can trust it to act — approval design, data handling and price. If you adopt one, start with read-only access, make the approval-required actions explicit, and widen scope gradually.
Sources
- Meta「Introducing Muse」 https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/
- Meta AI Muse https://ai.meta.com/muse/
- WIRED「The Battle to Be Your Personal AI Agent」 https://www.wired.com/story/ai-agents-dots-devday-muse-battling-it-out/
- WIRED「OpenAI Dots」 https://www.wired.com/story/openai-dots-always-on-ai-agents-that-proactively-help/
- CNBC「Meta pushes into personal AI agents」 https://www.cnbc.com/2026/09/08/meta-personal-ai-agents-public-reckoning-privacy-safety.html
- Reuters「Meta testing a human concierge」 https://www.reuters.com/business/meta-testing-human-concierge-its-new-personal-ai-agent-muse-2026-09-22/