AI news5
Google starts rolling out Gemini Omni, its video-generation answer to Sora
Google has begun rolling out Gemini Omni Flash, the first model in a new Omni family that generates and edits video from any mix of text, image, audio, and video input. It is reaching the Gemini app, Google Flow, and YouTube Shorts, with developer and enterprise API access to follow in the coming weeks. Flash-tier clips are capped at 10 seconds at launch, and every output carries an invisible SynthID watermark. Edits work through conversation, with characters and scenes staying consistent across instructions.
blog.google ↗Why it matters
Google is moving generative video from a standalone novelty into its mainstream consumer apps, putting direct pressure on OpenAI's Sora.
NVIDIA ships Vera, its first CPU built for AI agents, to the top labs
NVIDIA's Vera CPU has started arriving at major AI labs, with the first units delivered to Anthropic, OpenAI, and SpaceXAI, followed by Oracle Cloud. Vera packs 88 custom Olympus cores, 1.2 TB/s of memory bandwidth, and runs roughly 50 percent faster per core than a traditional server CPU. A single Vera rack links 256 of the chips to sustain more than 22,500 independent agent environments at once. It is the CPU half of NVIDIA's new Vera Rubin platform.
nvidianews.nvidia.com ↗Why it matters
agentic workloads lean heavily on CPU work like tool calls and orchestration, so purpose-built silicon signals how central agents have become to the AI stack.
OpenAI adds invisible watermarks and provenance metadata to every image it generates
OpenAI has joined the C2PA content-provenance standard and is embedding Google DeepMind's SynthID invisible watermark into all images made with ChatGPT, Codex, and the OpenAI API. C2PA metadata records where an image came from but can be stripped, while the SynthID signal lives in the pixels and survives screenshots and format changes. OpenAI also released a public verification tool, currently in preview, that lets anyone check whether an image came from its tools.
openai.com ↗Why it matters
durable provenance on AI images is becoming table stakes for trust, and it matters for any organisation that needs to disclose or detect AI-generated visuals.
Figure's humanoid robots run 200 hours nonstop and sort 250,000 packages
Figure AI ran three of its F.03 humanoid robots through a 200-hour continuous logistics test, sorting close to 250,000 packages with no hardware failures. The run was originally planned for 8 hours and went 25 times longer. The robots worked in shifts, rotating out to charge, and ran fully autonomously on Figure's onboard Helix-02 neural network with no teleoperation. Each package took roughly 2.6 to 2.8 seconds.
interestingengineering.com ↗Why it matters
sustained, failure-free autonomous operation is the threshold humanoids need to cross to move from demo to real shift work.
NVIDIA releases Nemotron-Labs-Diffusion, betting on faster non-autoregressive text
NVIDIA released Nemotron-Labs-Diffusion, a family of open models (3B, 8B, and 14B) that can switch between standard autoregressive generation, diffusion, and self-speculation decoding inside one architecture. In diffusion mode it denoises many tokens in parallel rather than one at a time. NVIDIA reports the 8B model decodes 5.9 times more tokens per forward pass than Qwen3-8B with better accuracy, for about 4 times higher throughput on its benchmark.
marktechpost.com ↗Why it matters
a major chipmaker treating diffusion text generation as production-ready, not a research curiosity, could reset expectations for inference speed and cost.
AI in the nonprofit sector5
Australian charity Uniting NSW.ACT saves 580 staff hours with an AI call agent
Uniting NSW.ACT, a community-services nonprofit with more than 12,000 staff, has been running an AI voice agent called Jeanie as the first point of contact for client inquiries, including homecare bookings. In its first six months Jeanie resolved more than 60 percent of routine inquiries on first contact and fully handled over 40 percent of incoming calls without a human. The charity reports about 580 hours of staff talk time saved, a 12 percent drop in total call volume, and 220 hours of client wait time removed.
blog.webex.com ↗Why it matters
it is a concrete, numbers-backed example of a large Australian nonprofit using AI on the front line of service delivery, not just in the back office.
Regis Aged Care uses AI to take paperwork off clinicians across 72 homes
Regis, one of Australia's largest aged-care providers, has deployed an assistant called RegiCare Assist, built on Microsoft Copilot Studio and Foundry with Cognizant. About 150 staff across 72 aged-care homes use it to handle progress notes and flag clinical concerns like pain, infection, agitation, and changes in condition. Staff report less time on admin, lower anxiety, and more time with residents.
news.microsoft.com ↗Why it matters
aged care is one of the clearest cases where cutting documentation load directly returns hours to frontline carers.
Mali nonprofit RobotsMALI uses AI to publish children's books in local languages
RobotsMALI, a nonprofit education initiative, uses an AI workflow to translate author-drafted stories into Malian languages such as Bambara, after Mali shifted away from French in 2023. In under a year it has produced more than 100 culturally grounded children's books and distributed them to over 300 pupils. AI handles translation and personalisation, while local authors and educators keep editorial control.
orfonline.org ↗Why it matters
it shows AI helping close a mother-tongue learning gap that commercial publishers have never served.
LawDroid releases a free, open-source Claude plugin for civil legal aid
LawDroid has launched a free, open-source Legal Aid Plugin for Anthropic's Claude, aimed at civil legal aid groups, court self-help centres, and public-interest law offices. It supports intake, eligibility screening, document drafting, attorney supervision, case handoffs, funder reporting, and client communication. Because it is open-source and needs no separate account or subscription, organisations can run it without new vendor costs.
lawnext.com ↗Why it matters
legal aid is chronically under-resourced, and a no-cost, Claude-based tool lowers the barrier for small offices to serve more clients.
Charity fits 26 visually impaired people with AI smart glasses in India
The Almawakeing Foundation, with Dr. Shroff's Charity Eye Hospital, has fitted 26 visually impaired people with AI-enabled smart glasses under its Enable Program. The glasses provide navigation, reading assistance, facial and currency recognition, surroundings detection, and an emergency SOS. The program is positioned as an accessibility and independence intervention rather than a clinical one.
hospitalmanagement.net ↗Why it matters
assistive AI hardware is reaching beneficiaries through charity programs, putting capability in the hands of people priced out of commercial devices.