Zuckerberg advances “personal superintelligence”: Meta releases its first personal agent, Muse, with mixed results from internal tests

On September 9, 2026, Meta officially launched its first personal intelligence agent, Muse (internal code name Hatch), aiming to achieve the “personal superintelligence” vision planned by CEO Mark Zuckerberg. This agent can independently perform tasks such as sending emails, selling cars, and booking trips on behalf of users. It is currently available only in the United States and can be used through the dedicated app or WhatsApp. Initially, it offers a free version and subscription plans costing $20 and $100 per month, with users able to opt out of data training. Muse runs on an independent cloud virtual machine based on the open-source model OpenClaw. Although it has built-in monitoring and censorship mechanisms, internal tests revealed issues with poor management when handling sensitive data access rights. Some employees reported serious vulnerabilities such as bypassing security restrictions, frequent forced logouts, and functional failures. Meta expects to invest over $130 billion in this area this year to diversify revenue and reduce its dependence on advertising.