the brief
Today skewed toward real shipping software and practical infrastructure. OpenAI brought ChatGPT Work and Codex to Linux, Vercel pushed Connect deeper into the CLI with new observability, Nvidia posted Nemotron 3.5 Lightning, and Modular hit Mojo 1.0. On the research side, reasoning traces leakage, context pruning for agents, and Cloudflare’s DDoS report all carry direct operational implications.
OpenAI released a Linux preview of its ChatGPT desktop, bringing ChatGPT Work and Codex to developers’ native workflows, though Computer Use remains constrained to the in‑app browser.
Now in preview: The ChatGPT desktop app for Linux. Use ChatGPT, ChatGPT Work, and Codex where you already work and build, with your projects and browser workflows on supported Linux systems.
You can now import and auto‑sync projects, chats, skills, and plugins into ChatGPT Work and Codex, consolidating tooling and history across agents inside the desktop app.
You can now keep your work from other agents in sync with ChatGPT Work and Codex. Import projects, chats, skills, and plugins, review your import history, and opt in to automatic updates in Settings. Now available in the ChatGPT desktop app.
Teams can create and attach connectors entirely from the terminal, with pre‑filled metadata, auth, and MCP or discovery URLs, streamlining agent and integration setup in CI.
Vercel Connect now supports CLI setup for 100+ connectors — You can now integrate 100+ services through Vercel Connect from the CLI. Previously, completed setup in the terminal for some services, and opened the dashboard for everything else.vercel connect create Pass the service name to create a connector, then attach it to your project: The CLI pre-populates the brand name, icon, auth type, and MCP or discovery URL, then prompts for any credentials the service requires and creates the connec...
Vercel now shows who created and used each connector token and when, with plan‑based retention and webhooks, making key management auditable across projects.
Vercel Connect adds observability support — Vercel Connect now gives teams line-level visibility into the token lifecycle. See who created a token, what app or project used it, when, and whether it is still active. Every connector's detail page includes a new tab:Observability Connect observability is available on all plans. Events are retained for 12 hours on Hobby, 3 days on Pro, and 30 days on Enterprise. To retain events longer, forward them to a custom webhook endpoint by adding a Drain ...
Vercel’s EMU centralizes identity and account lifecycle via SAML SSO and Directory Sync so organizations own employee accounts on company domains, tightening governance for Vercel at scale.
Enterprise Managed Users is now generally available — Enterprise Managed Users (EMU) gives organizations full control over the Vercel accounts tied to their verified domains. It makes the organization's identity provider the single source of truth for authentication and account lifecycle, so accounts on company domains are governed centrally rather than owned by individuals. Now generally available Requires an Enterprise plan, enforced SAML SSO, active Directory Sync, and at least one verifie...
An open 30B MoE model optimized for speed promises up to 4× faster output and ships on Hugging Face, targeting cost‑efficient assistants and agent routing.
Nvidia Nemotron 3.5 Lightning — Article URL: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 Comments URL: https://news.ycombinator.com/item?id=49257947 Points: 7 # Comments: 0
A $240M, multi‑year deal will deploy Nvidia HGX B300 on IBM Cloud to serve open‑source models at scale, signaling enterprise demand for dedicated inference capacity.
IBM and Together AI sign a $240M, multi-year deal to build an AI inference cluster on IBM Cloud, using Nvidia's HGX B300 systems, to support open-source models (Anhata Rooprai/Reuters) — Anhata Rooprai / Reuters: IBM and Together AI sign a $240M, multi-year deal to build an AI inference cluster on IBM Cloud, using Nvidia's HGX B300 systems, to support open-source models — IBM (IBM.N) and startup Together AI have signed a $240 million multi-year agreement to build a large-scale artificial inte...
The Python‑superset for high‑performance AI hits 1.0, aiming to simplify GPU‑accelerated kernels and ML systems programming with a production‑ready toolchain.
Mojo 1.0 Is Here — Article URL: https://www.modular.com/blog/modular-26-5-mojo-1-0-is-here Comments URL: https://news.ycombinator.com/item?id=49261128 Points: 43 # Comments: 13
The latest update resolves rare TUI redraw freezes, Git path detection on Windows, model reversion on /tui, and cross‑session inbox issues for a more stable IDE experience.
v2.1.228 — What's changed Fixed interactive sessions that could stop redrawing entirely, while the process kept running, after a rare internal layout error Fixed git / Git Bash not being found on Windows when Claude Code is launched from a parent folder of the git installation Fixed /tui reverting the session to an earlier model when /model had been changed since the last response Fixed cross-session messaging sometimes starting without an inbox in the first session after install or upgrade F...
New research shows Claude, GPT, and Gemini return encrypted chain‑of‑thought blocks that can be replayed to elicit plaintext traces from weaker same‑vendor models, raising serious data‑leak and eval‑integrity concerns.
Stealing Reasoning Traces from Proprietary LLM APIs — <p><strong><a href="https://stolen-thoughts.com/">Stealing Reasoning Traces from Proprietary LLM APIs</a></strong></p> A vanity domain name (<code>stolen-thoughts.com</code>) for <a href="https://www.alphaxiv.org/abs/2608.09867">a neat paper</a>:</p> <blockquote> <p>Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients that can be replayed across sessions, users, and models. We take a trace produced by a frontie...
Google reports its multimodal medical agent achieving near‑expert performance in simulated audio‑video clinical visits, a step toward safer, context‑rich healthcare assistants.
Advancing AMIE towards expert-level audio-visual clinical consultations — Health & Bioscience
Cloudflare’s H1 2026 report highlights 1 Tbps‑scale events driven by DNS and CLDAP reflection amid geopolitical conflict, underscoring the need for hardened DNS and scrubbing capacity.
Cloudflare DDoS Threat Report H1 2026: 1 Tbps attacks soar as DNS floods and geopolitical tensions drive a new wave — In the first half of 2026, Cloudflare detected a 519% surge in hyper-volumetric DDos attacks across its network. These attacks were driven heavily by DNS and CLDAP reflection vectors. This report breaks down how major geopolitical conflicts reshaped the global cyber threat landscape.
A study maps where to trim intermediate context in deep‑research agents to slash token usage without hurting quality, offering practical heuristics for long‑context workflows.
A study reveals that pruning context in deep research agents can cut token usage by 73% while maintaining quality. It offers insights into where and how pruning should be applied for maximum efficiency in complex queries. https://arxiv.org/abs/2608.08389
Researchers propose TCF, converting free‑form generation into controlled responses to verify model provenance, reporting AUC 0.986 and a path to IP enforcement without white‑box access.
Researchers developed TCF, a black-box fingerprinting technique to verify large language model ownership, achieving AUC 0.9861. TCF redefines verification by converting open-ended text generation to controlled responses, enhancing AI intellectual property protection. https://arxiv.org/abs/2608.08195
IBM Research and Hugging Face detail token‑efficient structured reasoning that approaches ACE‑level gains with slimmer prompts, cutting costs for multi‑step agent plans.
Thinking of ACE? We Can Do It with Fewer Tokens
A GitHub engineer shows how three tweaks to dependabot.yml reduced PR spam on a large repo, making security updates manageable instead of ignored.
A dozen Dependabot pull requests on a Monday morning is how important updates get ignored. On Microsoft's GCToolkit, roughly one in six commits were single-dependency version bumps. Three small changes to dependabot.yml fixed it 👇
Sophie Alpert’s guidance for engineers using LLMs on writing stresses clarity, accountability, and human review, a useful template for teams formalizing AI‑assist norms.
There are no lossless transformations of natural-language text — <p><strong><a href="https://sophiebits.com/2026/06/25/there-are-no-lossless-transformations-of-natural-language-text">There are no lossless transformations of natural-language text</a></strong></p> Sophie Alpert shares her "internal policy on acceptable use of AI writing by engineers". It's a short read (supporting its own recommendations) and really good.</p> <p>If you chose to have LLMs help massage your writing the following ...
CDT argues chatbot outputs can receive First Amendment protection while leaving room for system‑level regulation, a useful framing as AI speech rules take shape.
AI chatbots raise a fundamental First Amendment question: Whose speech is it? CDT’s Becca Branum explains why chatbot outputs can receive constitutional protection while still leaving room for meaningful regulation of AI systems. Read the full report: cdt.org/wp-content/u...