AI CRITIQUE

Bir kişi, mavi ve beyaz arka plan üzerinde yüzünü kapatan büyük mavi bir yükleme çemberiyle birlikte bir akıllı telefon tutuyor.

Meta's Muse app focuses on user surveillance

Read the article English (automatic translation)

Summary

WIRED's hands-on review of Meta's Muse AI agent demonstrates that its web browsing and task automation capabilities are genuinely competent. However, the application's core design motivation is to collect personal data, such as bank accounts, passports, and inbox contents. Furthermore, the app automatically enrolls users in AI training.

Why it matters

Muse received over 900,000 downloads in its first week and is being cross-promoted via Instagram, WhatsApp, and Messenger. This means Meta's data collection scope has expanded to millions of users, many of whom may not realize that their interactions are feeding model training by default. The fact that the opt-out configuration is enabled by default and the persistent Memory feature cannot be fully disabled sets a precedent for how mainstream AI agents manage sensitive personal data at scale.

Highlights

Virtual machine-based web browsing represents a real advancement. Muse navigated a bakery site, managed a single-choice cheese restriction, and left a note for the kitchen, accomplishing all of this with minimal errors.
Users are automatically included in AI training. The toggle to disable this feature is hidden under 'Data controls,' while the Memory feature has no off switch at all.
Meta's own security blog acknowledges that agent actions (such as reservations and Marketplace purchases) can 'indirectly influence' the ads users see. This blurs the line between a personal assistant and advertising signals.

Critical take

The article's strongest claim is that Muse is 'obsessed with data collection rather than getting things done.' This assertion is partly based on the reviewer's short testing period and the app's aggressive onboarding, which may reflect a launch-phase growth strategy rather than a permanent architectural flaw. On the other hand, Meta's claims regarding its 'private' virtual machine and data being 'cleaned' before training are accepted in the article but largely left unexamined, making the privacy critique one-sided.

Bu kritik, yerel yapay zeka modeli Qwen3.8-27B tarafından yazılmıştır. Haberi ben seçtim; kritik metnini yayımlanmadan önce tek tek denetlemiyorum.This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.

Back to the issue