Curated AI news from the last 24 hours — models, agents, enterprise, and developer tools.
Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full, its first frontier models and successors to the lightweight Muse Spark. Meta says they score comparably to leading rival models on benchmarks and lead on MemoryBench. Muse Wonder is available via the Meta Model API at $4/$20 per million input/output tokens, while Wonder Full is limited to Muse Code subscribers.
techcrunch.com ↗Satya Nadella wrote that it is time to reassess the trust architecture of AI, proposing that models be separated from the systems orchestrating them, actions be logged, and an authorized person be able to halt a model mid-task. The remarks follow reports of AI companies losing control of models and Anthropic CEO Dario Amodei's call for more cautious development.
techcrunch.com ↗Microsoft's president argued that the UK is best positioned to lead on AI regulation. The comments add to the ongoing policy debate among major AI vendors over how frontier models should be governed.
techcrunch.com ↗CNBC reports on Google's Gemini 4 arrival and investor focus shifting toward a breakout personal AI agent. Access to the newest model is initially restricted to select cybersecurity defenders and enterprise cloud customers.
cnbc.com ↗Apple disclosed a deal to hire the team and license technology from Huxe, a startup focused on personalized AI-generated podcasts. The arrangement follows an acqui-hire-style structure rather than a full acquisition.
techcrunch.com ↗TechCrunch examines whether dropping non-disclosure agreements around data center projects can ease community opposition. The piece looks at transparency as a response to growing local resistance to AI infrastructure buildouts.
techcrunch.com ↗A startup claims its AI can produce permit-ready architectural blueprints within minutes. The report covers an industry application of generative AI in construction and design workflows.
techcrunch.com ↗