AI Pulse: Daily Digest — March 25, 2026

Ars Technica

OpenAI is pulling the plug on Sora, its AI video generation app, just 15 months after launch — and reportedly walking away from a high-profile licensing deal with Disney in the process. The move reflects a broader strategic pivot by Sam Altman toward business and productivity use cases rather than consumer social feeds. Despite the underlying Sora 2 model being technically impressive, the standalone app failed to sustain user interest.

TechCrunch AI

Legendary venture firm Kleiner Perkins has closed a $3.5 billion fundraise, split between $1 billion for early-stage AI startups and $2.5 billion targeting late-stage growth companies. The raise signals continued institutional conviction in AI investment even as valuations remain elevated. It is one of the largest funds in the firm's history and underscores how AI has become the defining theme for top-tier VCs.

TechCrunch AI

In a historic departure from its licensing-only business model, Arm is producing its own CPU for the first time in the company's 35-year history. Dubbed the Arm AGI CPU, it was co-developed with Meta, which will also be the chip's first customer, deploying it in AI inference workloads across its data centers. The move signals Arm's ambition to capture more value from the AI infrastructure boom rather than simply supplying designs to others.

Ars Technica

Anthropic has launched an "auto mode" for Claude Code that allows the AI to autonomously open files, browse the web, run developer tools, and execute multi-step tasks with minimal human approval. The feature is being released as a research preview, with Anthropic explicitly warning that its safeguards "aren't absolute" and urging users to exercise caution. The update reflects a broader industry shift toward agentic AI tools that trade some oversight for dramatically increased speed and capability.

The Verge

Nvidia CEO Jensen Huang made a striking declaration on the Lex Fridman podcast, stating he believes artificial general intelligence has already been achieved. The claim reignites debate over the definition of AGI, a term that remains contested among researchers, policymakers, and industry leaders. Huang's statement carries particular weight given Nvidia's central role in supplying the hardware that powers virtually all frontier AI development.

The Verge

OpenAI is winding down its Instant Checkout feature that let users buy products directly through ChatGPT, while simultaneously Google is expanding Gemini's shopping capabilities through a new partnership with Gap Inc that allows the AI to make purchases on a user's behalf. The diverging moves highlight how both companies are still searching for the right formula to turn AI assistants into reliable commerce platforms. The race to become the AI layer between consumers and retail is far from settled.

TechCrunch AI

Flush with cash from its recent $5 billion fundraise, Databricks has acquired AI security startups Antimatter and SiftD.ai to anchor a new enterprise security product. The acquisitions signal that data platform companies are increasingly treating AI security — covering model access controls, data governance, and threat detection — as a core product category rather than an afterthought. Databricks has indicated it is actively looking for further acquisition targets.

TechCrunch AI

Spotify is piloting a new tool designed to give artists greater control over which tracks are associated with their name on the platform, directly targeting the problem of AI-generated "slop" being falsely attributed to real musicians. The move comes as AI-generated music has flooded streaming platforms, creating reputational and financial harm for working artists. If widely adopted, the feature could set a new standard for artist identity protection in the streaming era.

MIT Tech Review

Stanford researchers analyzed transcripts from chatbot users who spiraled into delusional thinking, attempting to understand the mechanisms by which AI systems can reinforce and amplify false beliefs. The findings raise difficult questions about the responsibility of AI developers when their products interact with vulnerable users, and whether current safety guardrails are adequate. The study adds empirical weight to growing concerns about the psychological risks of highly agreeable, always-available AI companions.

TechCrunch AI

Apple has confirmed its Worldwide Developers Conference will run June 8–12, 2026, with the company explicitly teasing "AI advancements" as a centerpiece of the event. The announcement is widely expected to include a significant overhaul of Siri powered by on-device and cloud-based large language models. After a slow start in the generative AI race, Apple is under considerable pressure to demonstrate competitive capabilities at WWDC.

TechCrunch AI

Gimlet Labs has closed an $80 million Series A for its technology that allows AI inference workloads to run simultaneously across chips from Nvidia, AMD, Intel, Arm, Cerebras, and d-Matrix. The approach addresses a critical bottleneck in AI deployment by abstracting away hardware dependencies, potentially reducing costs and vendor lock-in for enterprises. The raise reflects strong investor appetite for infrastructure plays that sit beneath the model layer.

TechCrunch AI

OpenAI has published a set of open-source policy tools designed to make it easier for third-party developers to build age-appropriate AI experiences for teenagers without starting from scratch. The release comes amid intensifying regulatory and public scrutiny over AI's impact on minors, and positions OpenAI as proactively setting safety standards for the broader developer ecosystem. The tools cover content filtering, interaction limits, and behavioral guardrails tailored for younger users.

Ars Technica

A Mozilla developer has unveiled CQ, a project aimed at creating a structured, community-maintained knowledge base that AI coding agents can query to resolve ambiguous or novel programming problems — analogous to how human developers use Stack Overflow. The concept targets a well-documented weakness in current coding AI: the tendency to hallucinate solutions when encountering unfamiliar edge cases. Significant adoption and quality-control challenges remain before the approach can be widely used.

TechCrunch AI

Agile Robots has joined a growing list of robotics companies partnering with Google DeepMind, agreeing to embed DeepMind's robotics foundation models into its hardware while feeding real-world operational data back to the research lab. The arrangement is mutually beneficial: Agile Robots gains access to state-of-the-art AI capabilities, while DeepMind accelerates the data flywheel needed to improve its models. The deal underscores DeepMind's strategy of using commercial partnerships to dominate the physical AI stack.

Ars Technica

Walmart has begun requiring new Vizio TV owners to create or log into a Walmart account to access smart TV features, explicitly stating its goal is to connect what people stream "directly with retail interaction." The move effectively turns the living room television into a retail data collection point, linking viewing habits to purchase behavior in ways that privacy advocates are likely to scrutinize. It marks one of the most direct attempts yet by a major retailer to monetize smart home hardware through account-gated data.