# Meta's Muse AI agent had a serious security flaw

> A zero-day exploit let attackers hijack the AI assistant by redirecting its processing to malicious servers.
- Title: Meta's Muse AI agent had a serious security flaw · The Verge
- Summary: A zero-day exploit let attackers hijack the AI assistant by redirecting its processing to malicious servers. Meta prioritized launching Muse quickly over…
- Keywords: meta, muse, ai agent, zero-day exploit, security vulnerability, patrick wardle, macos, privilege escalation, ai security, transcription processing, technology, culture
- Source: The Verge — https://theverge.com/tech/998679/meta-muse-patch-zero-day-exploit-ai-agent
- Author: Jess Weatherbed
- Published: 2026-09-22T11:53:58+00:00
- Read time: 1 min
- Topics: meta, muse, ai agent, zero-day exploit, security vulnerability, patrick wardle, macos, privilege escalation, ai security, transcription processing, technology, culture
## Meta patches Muse security flaw
Meta released a hotfix for its Muse macOS app after security researcher Patrick Wardle discovered a zero-day vulnerability that could let attackers take control of the AI agent.
## How the exploit worked
The vulnerability exploited an undocumented Muse setting that let attackers redirect transcription processing from Meta's servers to their own, gaining access to the user's Muse account.
## What attackers could do
Wardle's proof-of-concept attacks showed he could take pictures and write malicious files to disk through Muse without alerting the user, leveraging the AI agent's own privileges.

> We can manipulate the agent and leverage its privileges to do whatever we want.
## Design flaws enabled the attack
Muse processes dictation in the cloud instead of on-device, and allows any app to control all of its undocumented settings, both decisions that made the exploit possible.
## Wardle's criticism of Meta
The researcher argues Meta should have prioritized security from the start of Muse's development rather than treating it as an afterthought following launch.

> They should be thinking about security from the very start, and they are just not.
## Meta downplays the real-world risk
Meta says the exploit required local access to a user's device and malicious code already running, making the practical risk quite low despite the vulnerability's severity.
## Key takeaway

Meta prioritized launching Muse quickly over building security into its design from the start, a pattern security researchers say is common in AI development.