Meta’s Muse surges in popularity as security and platform disputes emerge
publishers 23articles 35first reported 23 Sep, 14:37 UTCdeveloping since 22 Sep · 8 editions
Meta’s new Muse AI assistant has climbed to the top of Apple’s App Store while drawing scrutiny over security, privacy and its ability to act on users’ behalf.
The agent can perform tasks including booking meetings and canceling flights, Business Insider reported, and completing authorized purchases, according to TechSpot. Meta also announced a forthcoming handheld “Muse Charm” that displays the assistant as a customizable mascot. CEO Mark Zuckerberg has said he expects Muse to become a “personal superintelligence” used by billions of people.
Business Insider reported that Muse was tested for months by tens of thousands of Meta employees under a team led by former GitHub CEO Nat Friedman. The outlet said the assistant had struggled with language coherence shortly before release but improved through extensive task training.
Meta has since patched a vulnerability in Muse’s macOS app discovered by security researcher Patrick Wardle. The flaw could let malicious software already running under a user’s account redirect cloud-based transcription processing and use Muse’s privileges to take actions such as capturing images or writing files. Meta said it issued a hotfix and characterized the practical risk as low because exploitation required local access rather than a remote attack.
Amazon has separately blocked Muse from shopping on its platform, saying the agent did not properly identify itself and posed security and privacy risks. Meta has said Muse cannot directly view users’ passwords or payment methods because credentials are kept in secure storage. Amazon is actively working to block other third-party agents, such as those from Google and OpenAI, according to TechSpot.
Muse’s tools also include identifying and canceling recurring subscriptions, potentially reducing charges sustained by consumer inattention or difficult cancellation processes.
HOW THIS STORY WAS MADE
Written from 8 articles, 4 with the publisher's own text; 8 independent newsrooms once syndicated copies count as one; this version written 42 h after the record first saw the story; the editor revised it.
- Publishers
- 23
- Source articles
- 35
- Given to the writer
- 8, 4 with the article's own text
- Independent newsrooms
- 8
- This version written
- 42 h after the record first saw the story
- Second model (editor)
- revised
- Publication gate
- passed
- Human review
- none
WHAT THE EDITOR DID
The second model checked the writer's draft against the same sources and changed it before publication. Struck words were in the draft; underlined words are its corrections.
WHAT IT FLAGGED editor revise: The story is mostly supported but needs narrower wording and attribution for Muse’s capabilities, testing, and Amazon restrictions. - "and subscriptions": The sources support Muse identifying and canceling recurring subscriptions, but do not source a general ability to cancel subscriptions as part of the same task list from Business Insider. - "The agent can perform tasks including booking meetings, canceling flights and subscriptions, and completing authorized purchases.": Completing authorized purchases is sourced to TechSpot, while booking meetings and canceling plane tickets are sourced to Business Insider; the sentence blends sourced details without attribution and overstates subscriptions as a general task. - "Business Insider reported that Muse’s launch followed months of testing by tens of thousands of Meta employees": Business Insider says tens of thousands of Meta employees tested Muse for months, not that the launch followed months of testing as a causal/timeline formulation. - "Amazon reportedly plans similar restrictions for agents from Google and OpenAI.": TechSpot says Amazon is actively working to block other third-party agents such as those from Google and OpenAI; it does not attribute this as a reported plan in the same wording.
THE DRAFT AND THE CHANGES
Meta’s Muse surges in popularity as security and platform disputes emerge
Meta’s new Muse AI assistant has climbed to the top of Apple’s App Store while drawing scrutiny over security, privacy and its ability to act on users’ behalf.
The agent can perform tasks including booking meetings,meetings and canceling flightsflights, andBusiness subscriptions,Insider reported, and completing authorized purchases.purchases, according to TechSpot. Meta also announced a forthcoming handheld “Muse Charm” that displays the assistant as a customizable mascot. CEO Mark Zuckerberg has said he expects Muse to become a “personal superintelligence” used by billions of people.
Business Insider reported that Muse’sMuse launchwas followedtested for months of testing by tens of thousands of Meta employees under a team led by former GitHub CEO Nat Friedman. The outlet said the assistant had struggled with language coherence shortly before release but improved through extensive task training.
Meta has since patched a vulnerability in Muse’s macOS app discovered by security researcher Patrick Wardle. The flaw could let malicious software already running under a user’s account redirect cloud-based transcription processing and use Muse’s privileges to take actions such as capturing images or writing files. Meta said it issued a hotfix and characterized the practical risk as low because exploitation required local access rather than a remote attack.
Amazon has separately blocked Muse from shopping on its platform, saying the agent did not properly identify itself and posed security and privacy risks. Meta has said Muse cannot directly view users’ passwords or payment methods because credentials are kept in secure storage. Amazon reportedlyis plansactively similarworking restrictionsto forblock agentsother third-party agents, such as those from Google and OpenAI.OpenAI, according to TechSpot.
Muse’s tools also include identifying and canceling recurring subscriptions, potentially reducing charges sustained by consumer inattention or difficult cancellation processes.
Written by a language model from the sources above, then checked by a second model that may only cut, attribute or correct. How it works →