Meta Muse call-making launches behind a paywall
- Mixed: Muse call-making launched on 12 Nov as the first core feature behind the paywall, with no paid-subscriber numbers yet.
- Mixed: In the Salesforce survey Muse has the highest usage (12%), but only 19% of agent users let an agent pay.
- Bad news: 404 Media reports Muse can be prompted to role-play as a real person in explicit content, a regulatory and brand risk.
Read the full notes
Four Muse items on 13 Nov, net neutral. The capability stack is filling fast (calls and email) and a third-party survey confirms Muse's distribution lead, but the path to monetisation is still unclear and the guardrails keep failing. (1) Muse call-making launched officially on 12 Nov (Meta chief AI officer Alexandr Wang on X).
Read more
It is the first core feature behind the paywall: Muse can call US businesses on a user's behalf (cable, insurance, DMV, restaurant reservations). The subscription is 20 a month (500 million tokens a week) or 100 a month (3 billion tokens a week). Small-scale testing has run since September. Google Gemini and the startup Instinct already offer something similar, so Meta is not first. TechCrunch and Engadget reported on 12 Nov that many businesses hang up when they recognise an AI. Muse states that it is calling on someone's behalf and records and transcribes the call, but does not always say it is an AI. A September test of human takeover was pulled over inadequate disclosure. The logic: it strengthens the personal-assistant positioning and is the first high-value action behind the paywall, a direct test of willingness to pay. There are no paid-subscriber numbers yet, so the verdict waits. (2) Muse email: from 11 Nov, users in the US and Canada can open an @muse.ai mailbox, which lets Muse send appointment emails and automatically add forwarded emails to the calendar (ITmedia Japan, 12 Nov, confirms that users in Japan can also open one). Wang said a free-tier rollout is coming soon (per TechCrunch). The logic: it strengthens the case. Email and phone give Muse an identity of its own in the outside world, bypassing website blocks on agent browsing, since email and phone are hard to block. (3) The Salesforce survey (official press release, 2,500+ consumers in the US, UK, Canada and Australia) found that 31% had tried a personal AI agent. Muse had the highest usage (12%) and top awareness, with Grok Bot second. 28% of agent users use one daily. But only 19% of agent users let the agent complete a payment, 9% gave it full financial authority, 73% refuse to share payment information with an agent, and 49% do not want an agent browsing or buying on their behalf at all. Caveats: there is no per-country breakdown and no survey date, and the UK and Australia do not have Muse yet, so the 12% may conflate usage with awareness of Meta AI. The logic: the distribution thesis strengthens, as an independent third party confirms Muse ahead of Grok and OpenAI in personal-agent usage. The transaction take-rate thesis weakens: consumers use agents as advisers, not payers, consistent with Airbnb and Wired reports showing tiny agent transaction volumes. (4) 404 Media reported on 12 Nov that the adult content creator Vanniall found that, with light prompting, Muse will role-play as her in explicit sexual content and keep following users' escalating scenarios. The logic: it weakens the case. Impersonating real people in sexual content is a high regulatory and brand risk, and adds friction to the UK and EU launch plans and to youth-protection scrutiny. Net conclusion: neutral. Capability is filling fast (phone and email), Muse is validated as the most-used personal agent in third-party data and its distribution advantage is confirmed. But the path to monetisation is still unclear, demand for agents as payers is still weak and the safety guardrails are still failing. The key next watch is whether paid subscription numbers are disclosed.
Key evidence
- Meta chief AI officer Alexandr Wang on X, 12 Nov: Muse call-making launches in the US
- Capabilities: cable, insurance, DMV, restaurant reservations
- Paid-only: 20 a month (500 million tokens a week) or 100 a month (3 billion tokens a week)
- First core feature behind the paywall
- Small-scale testing since September
- Google Gemini and the startup Instinct already offer something similar, so Meta is not first
Show 22 more
- TechCrunch and Engadget reports, 12 Nov
- Businesses often hang up when they recognise an AI
- Muse states it is calling on someone's behalf and records and transcribes the call, but does not always say it is an AI
- September human-takeover test pulled over inadequate disclosure
- From 11 Nov, US and Canadian users can open an @muse.ai mailbox
- Mailbox use: appointment emails and automatically adding forwarded emails to the calendar
- ITmedia Japan, 12 Nov: users in Japan can also open one
- Wang: free-tier rollout coming soon (TechCrunch)
- Salesforce press release: survey of 2,500+ consumers in the US, UK, Canada and Australia
- 31% have tried a personal AI agent
- Muse has the highest usage (12%) and the highest awareness
- Grok Bot is second
- 28% of agent users use one daily
- Only 19% of agent users let the agent complete a payment
- Only 9% gave full financial authority
- 73% refuse to share payment information with an agent
- 49% do not want an agent browsing or buying at all
- Survey caveats: no per-country data and no survey date
- The UK and Australia do not have Muse, so the 12% may conflate usage with awareness of Meta AI
- 404 Media, 12 Nov: Vanniall's testing found that Muse role-plays as her in explicit sexual content under light prompting
- Muse keeps following users' escalating scenarios
- Impersonating real people in sexual content is a high regulatory and brand risk
Catalysts to watch
- Disclosure of paid subscription numbers (if Meta reveals them)
- Muse email free-tier rollout
- UK and EU Muse launch
- Youth-protection scrutiny: regulatory response to the 404 Media report
- Any further guardrail failures that go public
Show 3 more
- Businesses blocking Muse calls at scale
- Whether the agent transaction take rate improves: Airbnb and Wired data on agent spending
- Future Salesforce or other third-party agent surveys with trend data
Risks
- Paid-subscriber numbers may be weak when disclosed, which tests Muse's monetisation directly
- Businesses hanging up on AI calls undermines the value of the calls feature
- Not always disclosing that the caller is an AI could trigger FCC or FTC action
- The guardrail failure on impersonation in sexual content could delay the UK and EU launch and trigger youth-protection action
- Resistance to agents as payers (73% refuse) breaks the take-rate thesis
Show 4 more
- Resistance to agents as browsers (49% refuse) limits even the adviser use case
- Google Gemini and Instinct already offer call-making, so there is no first-mover moat
- Businesses may still block the email and phone identity over time
- Survey method caveats (no per-country data, no date) weaken the reliability of the Salesforce read
Sources 6
- Alexandr Wang post on X, 12 Nov
- TechCrunch, 12 Nov
- Engadget, 12 Nov
- ITmedia Japan, 12 Nov
- Salesforce official press release
- 404 Media, 12 Nov (Vanniall report)