Meta's AI agent Muse has surpassed 10.2 million downloads in the United States within just 40 days of launch, a milestone reached by only 12 apps, according to TechCrunch. The fast start, however, has been fueled in part by heavy advertising spending, and fresh privacy concerns are now testing the product's momentum.

A Download Record Built on Marketing

Since its September launch, Muse has accumulated roughly 10.9 million downloads, with availability limited to the United States and Canada. Meta has leaned on the 3.6 billion daily users across its applications, combined with aggressive advertising, to drive adoption. According to Sensor Tower, Muse accounted for 39% of Meta's advertising budget in October, up from 23% in September, and climbed from 73rd to 21st in the US advertising-impressions ranking.

In its initial US phase, Muse outpaced ChatGPT, Grok, and Claude in downloads, although ChatGPT remains the global leader thanks to its broader reach.

Retention Is the Number That Matters

Download counts measure installs rather than sustained use, and with Muse's early growth so closely tied to advertising, engagement figures are the sharper test of whether that momentum reflects a lasting habit. Early engagement metrics look strong. Muse's seven-day retention reached 30%, compared with a 19% average among rivals; at the 14-day mark, the figures were 23% and 16%, respectively. Daily active users hit 2.5 million on October 16, up 15% week over week.

Citigroup projects that Muse could generate more than $27 billion in revenue by 2030, including roughly $23 billion from transactions and $4.5 billion from subscriptions. Those figures are projections, not actual sales. As previously reported by opolitan, Muse's launch coincided with an increase of more than $200 billion in Meta's market value. The central question is how many free users ultimately convert into paying customers.

A Sharing Flaw Undercuts the Privacy Pitch

Meta built Muse with security in mind, giving each agent its own secure virtual machine protected by a separate Sentinel agent that controls internet access. Yet Claroc researchers uncovered a flaw in Muse's privacy functions, finding more than 6,800 publicly discoverable links to shared conversations — some exposing personal budgets, internal financial reports, and confidential interview preparations. The exposure affected conversations users had chosen to share, not private chats.

In a statement to Ars Technica on October 20, Meta said it would suspend sharing, rendering the links tested by Claroc inaccessible. Whether the sharing feature returns, and under what safeguards, is now an open question.

Separately, Reuters revealed that Meta had tested human contractors handling Muse calls. One employee warned: "We are one bug away from unnecessary information being leaked to human callers." Meta subsequently rolled back the feature.

The Real Prize Is Agent-Driven Commerce

The opportunity extends well beyond Muse. As McKinsey has predicted, AI agents promise to transform marketing by helping people choose and purchase products rather than merely steering them toward advertisements.

The potential market is enormous. Morgan Stanley estimates that AI agents could influence up to $385 billion in US e-commerce by 2030, while Juniper Research expects annual agent-driven transactions worldwide to reach 120 billion by 2031. These are industry-wide forecasts, not projected revenue for Muse.

Muse is already moving into this territory through Stripe's Link checkout, which uses one-time card numbers to protect payment details.

For Meta, OpenAI, Google, and Anthropic, the competition is no longer just about attracting users. It is also about convincing people to trust AI with their purchases — and every privacy lapse makes that trust harder to earn.