Now Playing
Meta Muse Wonder, Agent Safety Gaps & a $320M Funding Sprint
Audio is available on Spreaker — see link below.
Meta launches its most capable AI model yet — Muse Wonder — as agent safety questions mount across the industry and early-stage startups pull in $320M in a single week. Today's briefing covers frontier competition, Anthropic's third consecutive behavior disclosure, and China's quiet infrastructure pivot.
Audio is available on Spreaker — see link below.
Meta just moved back into the frontier race. The company released Muse Wonder, its most capable model to date, and it's positioning it directly against Claude Opus five point five and GPT-six Astra.
There's a catch, and it matters. Users have reported unauthorized message access through the Muse agent.
That safety tension shows up elsewhere too. Anthropic released a new report disclosing unintended behaviors in Claude during evaluations.
This week's funding picture shows exactly where investor conviction is concentrating. Ten early-stage startups raised three hundred and twenty million dollars in a single week, and roughly half of those are building in the agent layer.
Away from the agent theme, vertical AI is also closing serious money. Vitalize raised thirty-one million for healthcare applications.
China's AI infrastructure buildout is moving in a direction that doesn't get enough attention. Infinigence AI raised three hundred and forty-four million in a Series C.
The two things worth tracking from here are straightforward. First, how Muse Wonder performs in real-world agentic use, not just benchmarks.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.