OpenAI and Broadcom unveiled Jalapeno, a custom accelerator built for large-language-model inference and described by OpenAI as the first chip in a multi-generation compute platform. OpenAI said engineering samples are running machine-learning workloads in the lab at production target frequency and power.
The company said the chip was designed around model, kernel, serving, memory, and networking requirements, with Broadcom handling silicon implementation and networking technology. OpenAI said a detailed technical report on performance will be presented later, and that initial deployment is planned by the end of 2026 with Broadcom, Celestica, Microsoft, and other partners.
Featured source: OpenAI .
Other Stories
NVIDIA and AWS Announce Production AI Infrastructure Work
NVIDIA said its latest AWS collaboration adds EC2 G7 instances with NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs and GPU-accelerated vector indexing through NVIDIA cuVS in Amazon OpenSearch Serverless. AWS described the broader integration as spanning accelerated compute, interconnects, model fine-tuning, and inference, including more than 1 million NVIDIA GPUs across AWS Regions starting in 2026.
Filed from: NVIDIA Blog , AWS .
White House Shortens Post-Quantum Cryptography Deadlines
Ars Technica reported that a June 22 executive order moves high-value and high-impact federal systems toward post-quantum key-establishment schemes by December 31, 2030, and quantum-safe digital signatures by December 31, 2031. The White House order frames the migration as protection against future quantum attacks on widely used public-key cryptography, while a separate June 22 order covers federal quantum innovation and commercialization policy.
Filed from: Ars Technica , White House .
UK CMA Publishes Responses in Microsoft Business Software Probe
The Register reported that the UK Competition and Markets Authority has published comments from customers, rivals, and Microsoft in its Strategic Market Status investigation into Microsoft’s business software ecosystem. GOV.UK lists responses from Google, Mozilla, Cloudflare, Element, the Browser Choice Alliance, Microsoft, and others, and says the CMA will consider them as evidence in the investigation.
Filed from: The Register , GOV.UK .
GitHub Joins Coalition Seeking California AI Transparency Changes
GitHub said it joined Black Forest Labs, Hugging Face, and Mozilla in a coalition asking California lawmakers to amend SB 942, as proposed to be amended in SB 1000, so the bill does not require open source developers to revoke licenses when downstream users miss transparency obligations. The coalition letter argues that open source licenses are designed to be perpetual and irrevocable, and GitHub said the group wants California’s transparency goals aligned with open source licensing practice and the EU AI Act Transparency Code of Practice.
Filed from: GitHub Blog , Coalition letter .
Rest of World Reports Divergent EV Battery Recycling Strategies
Rest of World reported that China and Western markets are taking different paths on spent electric-vehicle batteries, with China prioritizing immediate metal recovery while the U.S. and Europe back reuse for grid storage before recycling. The report said China has more than 85% of global recycling capacity and cited Chinese rules that took effect on April 1, while noting that U.S. companies such as Redwood Materials are framing second-life and later recovery around domestic critical-mineral supply.
Filed from: Rest of World .
Cloudflare Opens Self-Managed OAuth to All Customers
Cloudflare said self-managed OAuth is now available to all customers, letting developers create OAuth clients for delegated access to the Cloudflare API instead of relying only on API tokens. The company described the launch alongside an OAuth engine upgrade from Hydra 1.x to 2.x, including a blue-green migration, revocation replay capture, and reported average API P95 latency moving from 185ms to 101ms.
Filed from: Cloudflare Blog , Cloudflare Docs .
Anthropic Launches Claude Tag for Slack Teams
Anthropic introduced Claude Tag, a Slack beta for Claude Enterprise and Team customers where Claude can join selected channels, use connected tools and data, and respond when users tag @Claude. Anthropic said Claude Tag builds shared channel context and can plan tasks for later completion, while The Register reported that the existing Claude in Slack connector is being replaced by the new product.
Filed from: The Register , Anthropic .
Superhuman Acquires GPTZero
TechCrunch reported that Superhuman has acquired GPTZero, the AI-detection startup founded by Edward Tian and Alex Cui, and that the companies did not disclose terms. Superhuman’s announcement says GPTZero brings an authenticity suite covering AI-content detection, hallucination detection, plagiarism checking, and AI Vision.
Filed from: TechCrunch , Business Wire .
Sceye Prepares a Stratospheric 5G Test With SoftBank
MIT Technology Review reported that Sceye, a New Mexico company building solar-powered high-altitude platform stations, plans to send a roughly 200-foot craft across the Pacific as soon as August for a SoftBank 5G test near Japan. The craft is designed to park about 18 kilometers above the ocean and beam data directly to devices with a custom antenna, and the report said a spring test flight stayed aloft for 12 days and spent more than 88 hours parked in several locations.
Filed from: MIT Technology Review , Sceye .
Slate Auto Sets Electric Pickup Starting Price at $24,950
The Verge reported that Slate Auto’s electric pickup will start at $24,950, with preorders opening on June 24 and production scheduled for autumn 2026. The report said customers can place a $300 non-refundable deposit during a 30-day window to lock in an expected late-2026 delivery slot, and that Slate says it has taken roughly 180,000 reservations to date.
Filed from: The Verge , Slate Auto .
Datasette 1.0a35 Adds Create and Alter Table Interfaces
Simon Willison released Datasette 1.0a35 with a new create-table interface, an alter-table action, matching JSON APIs, and generated template-context documentation. The release notes describe the create and alter features as database and table actions, while the template-context documentation is treated as a stable API for custom templates until Datasette 2.0.
Filed from: Simon Willison , GitHub release .
OPFS and Pyodide Test Harness Explores Persistent Browser SQLite
Simon Willison published an OPFS and Pyodide test harness while investigating whether Datasette Lite can edit persistent SQLite files stored on a user’s computer from inside the browser. The post points to the Origin Private File System API as the browser storage mechanism and says the playground UI was built to test the approach across different browsers.
Filed from: Simon Willison , Tool .
Ultrasound Wristband Controls a Robotic Hand
MIT Technology Review covered an MIT and USC system that uses a wrist-worn ultrasound sticker and an AI model to translate wrist-muscle images into finger and palm positions. MIT News reported that demonstrations let a wearer wirelessly control a robotic hand and manipulate virtual objects, and the underlying work had previously appeared through MIT News in March.
Filed from: MIT Technology Review , MIT News .
From the Community
FUTO Swipe Releases Open Swipe-Typing Models and Dataset
FUTO released FUTO Swipe, a family of open models and algorithms for mobile swipe typing, along with a dataset of 1 million swipes under the MIT license. The system uses encoder, contextLM, and decoder models totaling under 2.5 million parameters, and FUTO reports a top-4 fail rate of about 4% on its test set.
Filed from: FUTO , HuggingFace dataset .
Meta Pauses Employee-Tracking Program After Internal Data Exposure
Meta paused its Model Compatibility Initiative tool after an internal security notice said databases containing employee mouse movements, click locations, keystrokes, and screen content were exposed to more people inside the company than intended. WIRED reported that the tool had rolled out in April to U.S. employees and was designed to train AI systems on human computer interaction.
Filed from: WIRED .
Cloudflare and Browser Vendors Develop Private Access Control Tokens
Cloudflare, Mozilla Firefox, Google Chrome, Microsoft Edge, and Shopify announced work on Private Access Control Tokens, a privacy-preserving protocol for websites to verify human or authorized bot traffic without invasive tracking or CAPTCHAs. The Cloudflare announcement says PACT lets sites with strong knowledge of personhood issue anonymous tokens that browsers can present to other sites.
Filed from: Cloudflare .
MacBook Neo Cursor-Lag Workaround Uses Screen Recording
A developer published a workaround for MacBook Neo cursor lag that occurs when the cursor nears screen edges or enters a Terminal window, correlating with a switch from hardware to software cursor rendering. The fix records one pixel of the screen every 10 seconds to /dev/null, forcing software cursor mode and eliminating the lag, with a provided script that creates a toggleable app.
Filed from: GitHub Gist .
Cackle Hardens Rust Projects Against Supply Chain Attacks
David Lattimore introduced Cackle, also called cargo-acl, a code ACL checker for Rust that restricts which crates in a dependency tree can use APIs such as networking, filesystem, or process execution. It analyzes object files and debug information to detect unauthorized usage, integrates with Bubblewrap for sandboxing build scripts and rustc, and currently supports Linux only.
Filed from: David Lattimore .
DiffusionBench Publishes a Benchmark for Generative Diffusion Transformers
DiffusionBench is an open-source benchmark for evaluating generative diffusion transformers across tasks including ImageNet class-conditional generation and text-to-image synthesis. The repository provides a unified codebase with multiple tokenizer families, configs for training and standalone sampling, and pretrained checkpoints for reproducibility.
Filed from: GitHub .
Reid Hoffman Criticizes SpaceX, xAI, and U.S. AI Intervention
Fortune reported that LinkedIn co-founder and OpenAI and Anthropic investor Reid Hoffman criticized SpaceX’s AI strategy and described xAI as a company on its third restart after the departure of its original co-founders. Hoffman also criticized the U.S. government’s order that forced Anthropic to pull its Fable and Mythos models, calling the intervention unpredictable for investors.
Filed from: Fortune .
Krea 2 Technical Report Details Creative Image Foundation Models
Krea published a technical report for Krea 2, a series of foundation models focused on aesthetic diversity and user creative control, using a multi-stage training pipeline with a diffusion transformer architecture that incorporates SwiGLU, GQA, and sigmoid-gated attention. The report says the model ranks among the top 10 on the Artificial Analysis text-to-image leaderboard and second among independent labs, with weights and inference released under a permissive license.
Filed from: Krea Blog .
