Muse and Instinct: Popular Success, Security Under Strain

Meta propels its agent Muse to the top of downloads and claims an isolated architecture for data. Instinct, currently raising $2.5 billion, is banking on Apple’s messaging. Both platforms draw inspiration from OpenClaw, whose meteoric rise triggered a wave of investments, but security guarantees remain incomplete.
Security: Isolated VM at Muse, vulnerabilities and possible access for Meta
OpenClaw faced significant security issues: one of its most downloaded skills contained malware, and according to a researcher, 15% of the skill directory had malicious instructions aimed at secretly accessing data or performing other suspicious actions. To address these risks, Mark Zuckerberg claims that Muse has been designed to ensure privacy and security, with data and identifiers stored in Muse Secure VM, an isolated Linux computer equipped with a browser, CPU, memory, and storage. However, the current isolation does not prevent Meta from accessing user data; the company plans to introduce a cryptographic and verifiable mechanism this year to prevent such access. By default, data is used to train and improve models, with an option for withdrawal. A researcher reported this week a zero-day vulnerability allowing the hijacking of the Muse agent and complete control via a simple attack. Instinct has also faced criticism for terms of use deemed too broad, which the company seems to have adjusted since.
Adoption: Muse tops the App Store with an estimated 600,000 DAU
According to Apptopia, Muse has climbed to the top of the App Store rankings and reportedly has 600,000 daily active users in the United States. The pricing offered by these two platforms is lower than that of Google Spark, while Instinct remains in beta. The one-click download of Muse has reduced the friction encountered with OpenClaw. Some users appreciate Muse's integration with Meta's systems. Instinct allows users to interact with AI agents via Apple’s default messaging, broadening the potential user base. Both solutions can operate on dedicated local devices, which presents potential security advantages, but this is not the default mode offered.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Lineage: Copying accusations, Meta's denial, and acknowledged inspiration
Internet users on social media have claimed that Muse is directly based on OpenClaw, citing the use of identical file names such as SOUL.md, similarities in documentation, and a similar interface. A Reddit user described Muse as a wrapper application built on OpenClaw, mentioning similar security risks. Meta has denied these allegations: Nat Friedman, product head of Superintelligence Labs, asserts that Muse was built from scratch while acknowledging strong product inspiration from OpenClaw. He states that he discovered OpenClaw in January and then purchased hundreds of Mac Minis for his team, aiming to create a comparable but safe, secure, simple, and scalable platform for billions of users. He clarifies that the team adopted some names and phrases that Peter Steinberger had, according to them, perfectly chosen. The influence of OpenClaw is thus acknowledged, even as Meta defends a complete technical overhaul.
Origins of OpenClaw: 10 months of iteration and global excitement
OpenClaw, initially conceived as a weekend project by a single person and designed to run on users' personal computers, transformed in ten months from a simple proof of concept into a tool considered useful. The agent offered the ability to communicate via WhatsApp, Telegram, Slack, Teams, or Discord. In about a week, it attracted two million visitors and garnered 100,000 stars on GitHub. Users purchased Mac Minis to run their agents continuously, a social network was created to allow AI agents to "chat," and meetups were organized worldwide. At that time, major AI companies were already betting on agent technology, but no product had achieved such success.
Industry timeline: from hiring Steinberger to Muse
In February, OpenAI hired Peter Steinberger, the creator of OpenClaw, to work on agents. In May, Google presented its ambitions in this area at its annual event, followed by Apple's commitment in June. In August, Instinct experienced significant growth in private beta and raised hundreds of millions of dollars while seeking funding at a valuation of $2.5 billion. In September, Meta unveiled Muse, its public agent, now available and gaining traction on the App Store.
Claimed uses of Instinct: from administrative tasks to travel
Instinct allows users to interact with their AI agents directly via messaging. Some report using it to fill out medical forms, cancel subscriptions, organize bachelor parties, book activities for vacations, obtain DMV appointments, or settle overdue toll bills. Others mention retrieving forgotten items, such as in a hotel in Canada.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.