Zero‑Day Bug in Meta’s Muse AI Allows Malware to Seize macOS Voice Assistant

Researchers have identified a critical zero‑day flaw in Meta’s Muse AI assistant for macOS that could let existing malicious code on a machine seize control of the voice‑activated tool, capture user commands, and steal authentication credentials.

The investigators who analyzed the defect observed that when malware runs with the same user rights as the genuine app, it can inject code into Muse’s process. This injection lets the attacker modify how the assistant handles prompts, converting the AI into a channel for additional malicious directives while staying hidden from the user.

The problem is especially concerning given that Muse is promoted as a productivity aid that listens for spoken queries and can open files, send messages, or fetch passwords on command. If an attacker commandeers the assistant, they could record sensitive credentials spoken to it, steer it to fetch further payloads, or alter its responses to deceive the user into divulging additional data.

Meta has not issued a public statement about the flaw, and its bug‑bounty program has not revealed any related submissions. The vulnerability was initially reported by cybersecuritynews, which emphasized how readily current malware could exploit the issue without needing privileges higher than those of the logged‑in user.

Experts note that the bug highlights a wider danger tied to AI‑powered assistants on personal computers. Although such utilities provide convenience, they also broaden the attack surface, particularly when they are deeply integrated with the system. Users should ensure macOS and all software are up to date, watch for abnormal activity from voice assistants, and think about restricting microphone permissions for apps that do not require them.

Meta is anticipated to roll out a fix in a forthcoming software update, though no exact schedule has been provided. Meanwhile, security specialists advise using endpoint protection capable of spotting irregular process injections and limiting the execution of unsigned binaries. As AI assistants gain traction, the sector is expected to scrutinize their security designs more closely to block comparable exploitation paths going forward.