Security researcher Patrick Wardle revealed on September 21 that malware already running on a Mac can quietly hijack Meta’s Muse AI assistant by altering a hidden setting. The proof‑of‑concept demonstrates how a malicious app can redirect voice input from the user to an attacker’s server instead of Meta’s servers.
When a user taps the microphone icon to dictate a prompt, the words are normally sent to Meta’s cloud for processing. Wardle’s demonstration shows that by changing a concealed configuration, the same audio is transmitted to an attacker’s endpoint, giving the attacker broad access to the assistant’s capabilities.
The flaw lies in Muse’s hidden setting, which allows the assistant’s microphone input to be rerouted. By exploiting this, malware can effectively turn the AI assistant into a covert backdoor.
Wardle’s proof‑of‑concept underscores the importance of scrutinising the permissions and hidden configurations of AI assistants, especially those that run on personal devices with extensive access rights.
While the demonstration was limited to a single Mac environment, it highlights a potential vector for attackers to intercept and manipulate voice commands in future releases of Meta’s Muse assistant.
<small>Source: The Hacker News — read the original story there.</small>
