Daily ·
Nvidia's 64GB DGX Spark; Cloudflare's Clef decision models — 2026-10-03 00:00
Ten stories: Nvidia brings a 64GB DGX Spark to market; Apple tightens macOS Full Disk Access over AI-agent risks; Cloudflare ships the open-weight Clef decision models; Black Forest Labs launches Flux 3 Image; a hands-on with OpenAI's Dots agent; Supabase raises $150M and buys Turso; Georgia holds an emergency meeting on AI exposing secret ballots; a $300M server-smuggling arrest; Suno adds a Speech feature; and a startup red-teaming models for psychological harm.
🎧 News Digest — 2026-10-03 00:00 UTC — 10 items
- Nvidia brings a 64GB DGX Spark to market this month — Nvidia's DGX Spark will be available with 64GB of unified memory from manufacturer partners Acer, ASUS, Dell, Gigabyte, HP and MSI. It keeps the same GB10 Grace Blackwell Superchip, DGX OS and full AI software stack as the 128GB model, supports up to 100-billion-parameter models on device, and two units can cluster via NVIDIA Sync Cluster Assistant to pool 128GB of memory and reach 200B parameters. Blender is among the first creator apps to support the platform. NVIDIA Blog · Oct 2
- Apple tightens macOS 'Full Disk Access' over AI-agent risks — Apple says it will add new controls around macOS's Full Disk Access permission, warning that increasingly capable AI agents make broad access to users' files, messages, mail and browsing history riskier. TechCrunch · Oct 2
- Cloudflare launches Clef and Clef-flash decision models — Cloudflare released two open-weight decision models that return brief classifications with probabilities instead of text. Clef is built on Qwen3.8-27B and Clef-flash on Qwen3.5-9B, both under the Apache 2.0 license, available on Hugging Face and running on Cloudflare's Workers AI. Cloudflare reports a median latency of about 39ms for Clef-flash and about 209ms for Clef, versus just over 524ms for TypeSafe AI's Jev, and positions the models against TypeSafe's Jev decision model. The Decoder · Oct 2
- Black Forest Labs launches Flux 3 Image — The new image model works like a canvas: you draw a bounding box for each element, describe what goes inside, and everything you did not touch stays exactly where it was. It is pitched for multi-step editing that leaves the rest of a picture alone. Black Forest Labs · Oct 2
- Hands-on: OpenAI's Dots agent — The Verge tested ChatGPT's new consumer agent, Dots. It says Dots feel less like a personal shopper and more like workplace software that happens to be able to order a burrito, with blobby, customizable avatars and a virtual machine the agent clicks around in. The Verge · Oct 2
- Supabase raises $150M and acquires Turso — Supabase, which commercializes the open-source PostgreSQL database, raised $150 million in a round led by Singapore's GIC sovereign wealth fund and joined by Alphabet's CapitalG, IronArc and SquarePeg. It also agreed to buy database startup Turso, whose SQLite-based database is optimized for AI agents, and launched Supabase Compute — hosted sandboxes for long-running AI agents. SiliconANGLE · Oct 2
- Georgia holds emergency meeting on AI exposing secret ballots — A Princeton researcher used a $20 subscription to an AI large language model and public election records to link voters to their ballots, recovering the voting order for about 1.52 million ballots — 98.9% of in-person ballots in 114 of the 139 counties examined. Georgia's state elections board held an emergency meeting days before early voting begins, and the outgoing secretary of state ordered tabulation IDs redacted from future public releases. The Guardian · Oct 2
- US arrests California man over $300M in smuggled computer servers — Greg Lui, 38, of California, was arrested and accused of smuggling more than $300 million worth of export-controlled computer servers to China, the Department of Justice announced. Prosecutors allege he used false paperwork and shipments through third countries including Malaysia and Singapore to re-export the gear to China from 2023 to 2024. He faces up to 50 years in prison if convicted. The Guardian · Oct 2
- Suno adds a 'Speech' feature — Suno is adding a feature called Speech to its AI music generator that creates spoken text with matching background music as a single audio track. Users type an idea or text and describe the voice and music style; the company pitches it for poems, meditations and bedtime stories. Suno has not said how the model was trained, as AI music generators face copyright litigation. The Decoder · Oct 2
- Circuit Breaker Labs red-teams models for psychological harm — Circuit Breaker Labs, one of TechCrunch's 2026 Startup Battlefield 200 finalists, builds AI agents it likens to "crash-test dummies" — simulations of users across ages, languages and cultures — to test whether models detect dangerous, psychologically harmful interactions. The startup works with human domain experts to build the simulations and uses a proprietary scoring method for auditable results. TechCrunch · Oct 2
🧪 Try This Next
- 🔧 Test Flux 3 Image's bounding-box workflow for storyboards and product shots — elements you don't touch stay put
- 🎥 Check Cloudflare's Clef models on Hugging Face if you build agents that need structured, low-latency decisions
- 📝 Save the Georgia ballot-privacy case — it is a template for every AI privacy fight ahead
More Dailies
Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.
