1. Google study: Nearly half of surveyed scientists use AI daily, saving almost seven hours a week even as untested hypotheses pile up Nearly half of scientists surveyed by Google and its research partners use some form of artificial intelligence every day, and respondents reported saving just under seven hours a week.

2. AI Agents Are Sending Emails and Carrying Wallets as 404 Media Documents a New Wave of Automated Nuisance An AI agent called Kudzu claimed it had read a 404 Media article, disagreed with it and independently emailed the technology publication to argue its case.

3. Vidu S2 pushes video generation into real-time interaction: 720p virtual characters can change outfits and scenes mid-conversation Vidu S2 is a video-generation system designed to turn one-off AI clips into streams that users can direct and edit as they run.

In Brief

• Google launches Gemini 3.8 Live voice models Google introduced Gemini 3.8 Live for cost-efficient voice agents and an Extended Thinking version for complex, multi-step tasks. Both are rolling out through the Gemini API and Google AI Studio, with varying availability across Search, Workspace, Gemini Live, and enterprise previews.

• Salesforce and Nvidia develop Koa reasoning model Salesforce’s first reasoning model is based on Nvidia’s open-weight Nemotron and post-trained with synthetic data for sales, marketing, and customer-support tasks. Salesforce plans to offer Koa through its Agentforce platform as an alternative to closed frontier models.

• Ninth Circuit lists Amazon–Perplexity case A U.S. Court of Appeals for the Ninth Circuit page lists a case between Amazon and Perplexity; the supplied candidate contains no details about the claims or disposition.

• Meta introduces AI-focused Meta One subscriptions Meta launched $7.99-per-month Core and $19.99-per-month Premium plans offering expanded access to image, video, voice, and editing tools across Facebook, Instagram, and WhatsApp. Separate business and creator tiers start at $14.99 per month and include varying levels of access to Meta Business Agent.

• WhatsApp Business gains an MCP server for AI agents Meta’s new WhatsApp Business Tools MCP connects coding agents such as Claude, Cursor, Codex, and ChatGPT to the WhatsApp Business Platform. Agents can create business accounts, register phone numbers for the Cloud API, manage templates, test messages and webhooks, and monitor configuration problems.

• AI-search marketing startup Profound raises $180 million Profound, whose software helps brands analyze and improve their visibility in AI-generated search results, raised a $180 million Series D led by Sequoia and Kleiner Perkins at a $1.8 billion valuation. The company says it has more than 1,000 enterprise customers and tripled revenue over the past six months.

• Perplexity deploys Astra across production work OpenAI says Perplexity uses its Astra model to draft communications, modify software, and monitor production systems end to end, with less frequent human check-ins than earlier models required.

• Forecast nearly doubles expected data-center gas demand BloombergNEF projects that U.S. data centers could consume about 18 billion cubic feet of natural gas per day by 2035, nearly twice its estimate from nine months earlier. The forecast includes 2.9–3.4 billion cubic feet per day for facilities generating power onsite and roughly 15 billion for grid-connected facilities.

• AIUC raises $40 million for independent agent audits Artificial Intelligence Underwriting Company raised a $40 million Series A for its AIUC-1 certification and testing service. The startup says it evaluates agents against roughly 5,000 scenarios involving jailbreaks, hallucinations, and data leaks, with humans verifying the final audit.

• PhysBrain 1.5 combines robot perception, action, and prediction PhysBrain 1.5 is an 8-billion-parameter model trained to understand physical environments, generate end-effector movements, and predict future visual states within one autoregressive framework. Its authors report a 72.5 average across 28 embodied-understanding benchmarks and the best open-source result on 14 of them.

• ZGCM-1 opens its compact model and full training pipeline ZGCM-1 is a fully open 7-billion-parameter model supporting mathematical reasoning, tool use, and contexts up to 256,000 tokens. Its developers released model weights, intermediate checkpoints, training code, data recipes, and logs, and report a 4.2-fold improvement in 16K pretraining time-to-loss.

Read the full edition →