Meta launches Muse, a 24/7 personal AI agent, plus Muse Spark 1.3 coding model

Per TechCrunch/TechXplore, Morningstar and Futurism, Muse runs continuously on cloud VMs with 'nightly reflection' and proactive planning, executing real-world tasks — email, travel, reservations, scheduling — rather than just chatting. Security relies on private VM isolation plus Llama Guard 3 and Prompt Guard, though a fully Confidential VM is only planned for late 2026. Meta also shipped Muse Spark 1.3 for coding and multi-agent workflows, positioned as comparable to GPT-5.6 Sol and Opus 5.

Markets loved it: Meta stock jumped 6.6% as the launch addressed a major investor concern — whether Meta could translate its AI spend into a consumer agent product. Alexandr Wang's 'blown past projections' framing reinforced the momentum narrative, and the launch caps what Meta calls a breakout AI year backed by heavy infrastructure investment, including a multi-billion-dollar Nebius Group deal.

But the community reaction was sharply skeptical. An HN thread (188 points, 180 comments) crystallized privacy fears — 'Why grant Meta access to all aspects of your life?' — with developers noting Meta can access data 'when necessary' despite Secure VM claims, and that the Confidential VM isn't ready. A Futurism investigation went further, contrasting the 'uber-creepy' chatbot with reports of child-abuse content flooding Meta platforms, framing the launch as tone-deaf.

Competitively, Muse is Meta's answer to autonomous agents from OpenAI (Operator-style), Anthropic and Google, differentiating on always-on operation and deep integration with Meta's recommendation systems for personalization. The core tension: the value proposition (an agent that knows everything about you) is also the risk. Watch enterprise/regulatory scrutiny, real autonomy limits, and whether the free-token generosity survives once demand is proven.