BREAKING
Rozrywka_od_emocji_do_wygranej_dzięki_the-magnetic-slots_com_pl_gdzie_szczęścCNC demands Codavaland autonomyGoogle Pixel 12 leak hints at major display and software upgradesVote Chori: INDIA Bloc Protest Was More About Optics and SymbolismUitgebreide_gids_voor_online_casinos_inclusief_een_blik_op_https_zombillioncasinProfessioneel_advies_rondom_the-westaces-nl_nl_zorgt_voor_optimale_resultatenMeta Rushed to Fix Serious Muse AI Security Flaws Before LaunchDefence Minister launches first indigenous FSS ‘Surya’ at HSLTemple Wearable Nears Launch; It uses PPG technologyNavaratri 2026: 9 Days, 9 Forms Of Maa Durga, Colours And Significance
Technology

Meta Rushed to Fix Serious Muse AI Security Flaws Before Launch

hellohimachalse@gmail.com 06 Oct 2026 1 hour ago

Meta reportedly discovered serious security vulnerabilities in Muse, its personal AI agent, just weeks before the product was launched. The issues were considered significant enough to trigger an urgent response, with employees working nights and weekends to strengthen the system.

According to a report by 404 Media, one of the vulnerabilities could potentially have allowed an ordinary Muse user to break out of the AI agent’s protected environment and gain access to sensitive information within Meta’s systems. The concerns were reportedly raised with CEO Mark Zuckerberg as the company brought multiple teams together to address the problems.

Muse runs in a protected environment

Muse is designed to act on behalf of users and can interact with services such as email, calendars, messaging platforms and other accounts. To reduce the risks associated with these capabilities, the AI agent operates inside a virtual machine, which is intended to keep it separated from Meta’s main infrastructure.

However, the report said one of the vulnerabilities was connected to an exploit discovered in Linux virtual machine code in July. If successfully exploited, the flaw could potentially allow an attacker to escape the virtual machine and reach the larger system hosting Muse.

The report, citing a Meta source and internal security documents, said at least one of the vulnerabilities could have given a regular Muse user access to sensitive data inside Meta’s systems. As more security concerns emerged, Meta reportedly brought together several teams to investigate and resolve them.

Engineers worked through weekends

The urgency of the situation was also reflected in an internal post from Meta executives dated September 18. The post reportedly said work on the issue began on August 27 and continued for several weeks, including weekends.

Engineers also focused on restricting Muse’s access. This included limiting the services and systems the AI agent could reach, as well as controlling its connections to the internet and Meta’s internal infrastructure.

The risks are particularly important because Muse is designed to do more than answer questions. As an AI agent, it can access connected accounts and perform tasks for users. A security breach could therefore have consequences beyond a typical chatbot conversation.

Meta says it has strengthened Muse’s security through extensive testing, internal security reviews and its bug bounty programme.

However, the latest concerns add to existing scrutiny surrounding the AI agent. Security researcher Patrick Wardle recently identified another vulnerability that could allow applications and terminal commands to control a user’s Muse. In another reported incident, a user managed to make Muse export Instagram followers, despite the feature not being expected to work that way.

As AI agents gain access to more services and perform increasingly complex tasks, security remains a key challenge. The Muse incidents highlight why strong isolation, continuous testing and quick responses are important before such systems are widely deployed.

Related News