today's updates show ai moving deeper into everyday tools and enterprise workflows. apple approved the first ai agent for its business messaging, nvidia released a model that handles safety across multiple modes, and meta launched a helper for facebook creators. meanwhile, new research warns about emotional dependence on ai, and benchmarks for voice agents expand into healthcare and it.

  1. apple approves poke as first ai agent on messages for business - this marks a shift in how businesses can use ai on apple's platform, letting customers text an agent for tasks like booking or support.
  2. nemotron 3.5 unifies multimodal safety with custom policies - the model checks text, images, and video in one call, supports many languages, and lets companies set their own safety rules, which could simplify content moderation.
  3. meta adds ai creator assistant to facebook - the assistant gives creators personalized tips on content and performance, potentially changing how they manage their pages.
  4. eva-bench grows to three enterprise voice agent domains - adding itsm and healthcare scenarios makes the benchmark more useful for testing voice agents in real business settings.
  5. ai emotional dependence grows from routine use, not just chatbots - the study found people can become emotionally reliant on ai through everyday tasks, not just companion apps, which raises concerns about long-term effects.

other notable news includes hello robot shipping stretch 4 for home research, hugging face rebuilding its cli for ai coding agents, and a new framework for verifying ai agents before deployment. ieee also proposed arithmetic standards for machine learning, and a guide showed how to fine-tune speech recognition for under-resourced languages.