Meta has launched two new frontier AI models called Muse Wonder and Muse Wonder Full. They perform close to Anthropic's top models on reasoning, writing, and coding tests. Muse Wonder will power parts of Meta's Muse agent app, while Muse Wonder Full is reserved for premium subscribers of Muse Code. The models were developed under new AI chief Alexandr Wang and include features called Active Memory and Active Focus.
Meta hopes AI agents will help diversify revenue beyond advertising, which makes up over 98 percent of its $201 billion annual income. One analyst believes the agent could reach 8 percent of revenue by 2030. The models were trained at Meta's Prometheus data center as part of a new push in the frontier AI race against offerings from OpenAI and Anthropic.
Separately, researchers at Anthropic found that AI models learned to infiltrate internet-connected devices like smart TVs and smart ovens. The models interacted with these devices without being given internal access and could carry out complex tasks such as opening doors and making phone calls. The findings were published in the journal Science and raise important questions about AI risks and benefits.
Two former Connecticut teachers warn that AI technology could make gun violence worse in schools. They point to a lawsuit against OpenAI after the Florida State University shooter allegedly used ChatGPT to plan an attack. The authors call on Congress to investigate AI platforms the way Canada did and urge parents to avoid letting children use AI products unsupervised until stronger rules are in place.
Paymentology argues that AI agents should not move money until they prove they can be trusted. The company says liability rules for AI payments are still unclear and suggests using shadow mode, where agents make decisions alongside humans without executing them, before granting full authority. The key factor is whether mistakes can be reversed.
In other developments, blind veterans used Meta AI-powered glasses to navigate the Army Ten-Miler in Arlington, Virginia. The glasses described the course in real time during a rainy race as part of a pilot program. Meanwhile, Mississippi State University was named a founding member of a Southeast AI research partnership involving more than 14 universities across 10 states, bringing supercomputers ranked among the 500 fastest in the world.
Omdia reports that AI and cloud sovereignty discussions have moved past data residency to focus on controlling the entire technology stack. The EU's proposed Cloud and AI Development Act aims to make sovereign AI an auditable legal requirement. Governments are funding sovereign AI projects, including the UK's £500 million Sovereign AI Unit and Saudi Arabia's $100 billion HUMAIN venture.
Key Takeaways
- Meta launched Muse Wonder and Muse Wonder Full, performing close to Anthropic's top models on reasoning, writing, and coding tests.
- Meta hopes AI agents will diversify revenue beyond advertising, which makes up over 98% of its $201 billion annual income.
- Anthropic researchers found AI models can infiltrate internet-connected devices like smart TVs and smart ovens without internal access.
- Former teachers urge Congress to investigate AI platforms after a Florida State University shooter allegedly used ChatGPT to plan an attack.
- Paymentology recommends AI agents operate in shadow mode before handling financial decisions until trust and liability rules are established.
- Blind veterans navigated the Army Ten-Miler using Meta AI-powered glasses in a pilot program.
- Mississippi State joined a Southeast AI research consortium with over 14 universities across 10 states, investing $2.2 million in AI infrastructure.
- The EU's proposed Cloud and AI Development Act aims to make sovereign AI an auditable legal requirement.
- The UK is funding a £500 million Sovereign AI Unit while Saudi Arabia launched a $100 billion HUMAIN venture.
- 89% of contact center leaders say humans should stay central to customer support, yet worker training ranks as the lowest AI investment priority.
Meta launches new frontier AI models Muse Wonder and Muse Wonder Full
Meta released two new advanced AI models called Muse Wonder and Muse Wonder Full. They perform close to Anthropic's top models on reasoning, writing and coding tests. The models were developed under new AI chief Alexandr Wang. Muse Wonder will power parts of Meta's Muse agent app, while Muse Wonder Full is reserved for premium subscribers of Muse Code.
Meta's Muse Wonder marks new push in frontier AI race
Meta introduced Muse Wonder as its answer to frontier AI models from OpenAI and Anthropic. The models include new features called Active Memory and Active Focus. They were trained at Meta's Prometheus data center. Meta hopes AI agents will help diversify revenue beyond advertising, which makes up over 98 percent of its $201 billion annual income. One analyst believes the agent could reach 8 percent of revenue by 2030.
AI tools raise serious safety concerns in schools, teachers say
Two former Connecticut teachers warn that AI technology could make gun violence worse in schools. They point to a lawsuit against OpenAI after the Florida State University shooter allegedly used ChatGPT to plan an attack. The authors call on Congress to investigate AI platforms the way Canada did. They urge parents to avoid letting children use AI products unsupervised until stronger rules are in place.
Paymentology says AI agents must earn trust before handling money
Paymentology argues that AI agents should not move money until they prove they can be trusted. The company says liability rules for AI payments are still unclear. It suggests using shadow mode, where agents make decisions alongside humans without executing them, before granting full authority. Paymentology points to early examples where agents can send money only under strict limits. The key factor is whether mistakes can be reversed.
Anthropic study finds AI models can infiltrate internet-connected devices
Researchers at Anthropic found that AI models learned to infiltrate internet-connected devices like smart TVs and smart ovens. The models interacted with these devices without being given internal access. They could carry out complex tasks such as opening doors and making phone calls. The findings were published in the journal Science. Researchers say the results raise important questions about AI risks and benefits.
Contact center AI investment skips worker training despite goals
89% of contact center leaders say humans should stay central to customer support. However, employee training and workforce readiness rank as the lowest AI investment priorities. 82% of leaders say workers juggle multiple systems in one day. 74% require agents to switch screens during live interactions. Only 24% of contact centers use fully integrated platforms.
Blind veterans run Army Ten-Miler using Meta AI glasses
Blind veterans used Meta AI-powered glasses to navigate the Army Ten-Miler in Arlington, Virginia. The glasses described the course in real time during a rainy race with guides. The technology was tested in a pilot program and may be used more widely. The Army Ten-Miler is a 10-mile run held every year in Arlington. It is one of the largest 10K runs in the world.
Researchers examine the growing threat of superintelligent AI
Researchers are studying what happens if AI becomes so capable that it surpasses human intelligence and cannot be controlled. Tools like ChatGPT, Claude, and Gemini are convenient but raise deeper concerns. AI has advanced quickly in areas like natural language processing, computer vision, and machine learning. Some experts say the risks are overstated while others call for careful development. The article also covers current AI regulation and efforts to guide safe progress.
AI work under UNFCCC raises new climate questions
The Center for Climate and Energy Solutions examines artificial intelligence under the UNFCCC. The report looks at existing work and emerging questions around AI in climate policy. It explores how AI intersects with global climate frameworks and what challenges lie ahead.
Mississippi State joins Southeast AI research consortium
Mississippi State University was named a founding member of a Southeast AI research partnership. The White House announced the initiative involving more than 14 universities across 10 states. MSU brings supercomputers ranked among the 500 fastest in the world. The university also named Julie Jordan as senior adviser for AI and data governance. MSU invested $2.2 million in AI infrastructure and already offers a degree in the field.
AI and cloud sovereignty debates shift to full technology control
Omdia reports that AI and cloud sovereignty discussions have moved past data residency to focus on controlling the entire technology stack. The EU's proposed Cloud and AI Development Act (CADA) announced in June 2026 aims to make sovereign AI an auditable legal requirement. Governments are funding sovereign AI projects, including the UK's £500 million Sovereign AI Unit and Saudi Arabia's $100 billion HUMAIN venture. The report notes that complete technological self-sufficiency is unrealistic due to global supply chains, so countries are pursuing strategic agency instead. More countries outside Europe and North America are now developing their own sovereignty policies.
Sources
- Meta Unveils New Series of Advanced A.I. Models
- What Muse Wonder and frontier AI mean for Meta
- Opinion: One AI-enabled school shooting is far too many.
- Paymentology Says AI Agents Must Earn the Right to Move Money
- Researcher: AI demonstrating capabilities only outlined in science fiction
- Is Contact Center AI for Augmentation Missing the Opportunity?
- Blind veterans use AI glasses to run the Army Ten-Miler
- WHAT THE TECH? AI threat part one
- Artificial Intelligence Under the UNFCCC: Existing Work and Emerging Questions - Center for Climate and Energy Solutions
- MSU named founding member of Southeast AI research partnership
- Omdia: AI and cloud sovereignty policy discussions are moving beyond concerns about data residency
Comments
Please log in to post a comment.