Science & Tech · AI · 1 hr ago
Meta details safeguards for Muse Wonder AI agents
Meta introduced its Muse Wonder models for AI agents that can work on tasks such as reviewing leases, preparing insurance claims and developing software.
The models are aimed at commercial and research use and can analyze text, images, video and audio.
They can carry details between conversations and try different approaches when a task hits an obstacle.
Because agents may access messages, calendars and private documents, Meta says the models are trained not to share private information without permission and to refuse harmful requests.
Meta says remaining safety risks are moderate or lower with safeguards, though it could not rule out high risk for chemical and biological misuse without them.
The company also acknowledges possible errors, bias and weaknesses, and recommends human confirmation before irreversible actions.
Muse Wonder is available through Meta’s Model API and Muse Code, while Wonder Full is available to Muse Code subscribers; Meta plans to bring Wonder to Muse, starting with subscribers.
Meta outlined safeguards for Muse Wonder and Muse Wonder Full, models designed to power AI agents that act on users’ behalf.
The models are intended for commercial and research use and can handle tasks including preparing insurance claims, reviewing leases and developing software.
Meta said the models are trained to protect private information, refuse requests that assist harmful acts and seek user confirmation before important actions.
The company assessed remaining risks as moderate or lower across cybersecurity, chemical and biological misuse, and loss of human control when safeguards are in place.
Meta acknowledged that the models can make errors and recommended human confirmation before irreversible actions.
- Who
- Meta developed the Muse Wonder and Muse Wonder Full models.
- What
- Meta outlined safeguards and capabilities for AI agents powered by the models.
- When
- The article was published on 2026-10-13.
- Where
- Not stated
- Why
- The models are designed to handle tasks on users’ behalf while protecting private information and reducing risks.
This story does not have two clearly opposing sides.
This story does not have a timeline yet.
- Models
- Muse Wonder and Muse Wonder Full
- Training location
- Meta’s Prometheus data center
- Risk assessment
- Moderate or lower across assessed areas with protections in place
- Availability
- Muse Wonder through the Meta Model API and Muse Code
- Availability
- Muse Wonder Full for Muse Code subscribers
What people are saying
Muse Wonder is our most aligned model to date, and alignment is central to how we build personal superintelligence. A full report with additional safety evaluations will be available soon
Coding brings these capabilities together. Building software requires tracking requirements, making changes across a codebase, and working through repeated cycles of testing and revision
Sources
Related news
Meta unveils two AI models it says rival leading systems
Anthropic says AI models may soon spread across home devices
Epic Games shifts toward AI to speed up game development
Alibaba Prepares Qwen Go! Agent for Claims, Refunds and Bookings
CEATEC opens in Japan with AI-powered robots and machinery in focus
Scientists call for international body to oversee AI safety
Saudi teams use AI and 3D printing to develop lower-cost disability technology
Microsoft president calls for enforceable global AI safety rules
UK sets out plans to use AI to strengthen cyber security
Anthropic says safeguards cut AI model extraction by up to 99%
CEATEC opens with AI for factories, health care and cybersecurity
