Meta has added another contender to the AI race by introducing ‘Muse’, a personal AI assistant that can pay your bills for you.
The AI assistant, downloadable as an app on mobile phones, can autonomously send emails, sell a car, and book travel on a person's behalf. The app is also accessible via WhatsApp and Instagram.
The idea has raised deep concerns, as an AI personal assistant would need to have access to sensitive personal data to function.
CEO Mark Zuckerberg said that the plan was to supply “personal superintelligence" to the billions of people who use Meta's services daily.
Muse can “open a browser, fill out forms, and negotiate” on behalf of the person using it.
During the rollout, Meta emphasised that the app is reliable as it runs on a Secure VM, a dedicated secure computer with its own browser. Meta says that Muse had no visibility into people's passwords or payment methods and that “credentials a person shares go into secure storage, so Muse can use them without seeing them, including passwords a person types into the browser themselves.”
Each user also stays in control of their “Muse” and can decide how much access it gets.
Meta also said that Muse can “remember what matters to a person, and they can always tell it to “forget” specific things it’s learned.”
The company is also introducing a different, more secure version of Muse later this year: Muse Confidential VM, where the whole VM, including a person’s data and conversations with Muse, is encrypted with a key only they hold, so not even Meta can access it.
The concerns come as Meta has had a history of privacy issues. Meta relies on advertisements for most of its revenue from Facebook and Instagram.
The company recently agreed to pay $17 million in a settlement with a coalition US state attorneys after it was sued for misrepresenting the prevalence of harm on apps like Facebook and Instagram.
The company also had to pay millions for its failure to warn the public about dangers its platforms posed to children.
There are also rising concerns about the cybersecurity risks posed by AI and its models;
Users of Muse have opted out of letting Meta use their interaction with Muse to train its AI models.
David Singleton, Meta vice president of engineering, said that if they don’t, the company will scrub “critical personally identifying information” before using the agent conversations and interactions to improve AI models.
The company is adding to the rising industry trend of personal AI models such as OpenClaw.
Meta employees who tested the tool reported mixed results according to internal posts accessed by Reuters.
One person said that Muse was useful, as it had become "the third participant" on a recent three-week honeymoon in Indonesia.
Others said that it had major security flaws. One time it routed around guardrails to expose a person's personal iCloud photos after being prompted to identify toys visible in pictures from a child's birthday party.
Meta has also been under fire for having trouble with its AI agents that bent rules and behaved in unpredictable ways, like OpenAI and Anthropic. In August, Meta’s attempt to replace workers with AI backfired after its technical and security incidents spiked upto 40 per cent from last year, according to a Reuters report.
Staff spent about 70 per cent more time firefighting the incidents.
