After a rollercoaster negotiation process with the Trump administration that dragged on for two weeks, Anthropic's Mythos 5 is finally back in action — at least, somewhat, for a select group of organizations, according to a letter from the government to Anthropic that was viewed by The Verge. Fable 5, however — the public-facing Mythos-class model — appears to still be in limbo, with no apparent timeline for a rollout agreement.

This guide explains how to build an MCP Apps host using @ai-sdk/mcp and @ai-sdk/react, including filtering model-visible tools, reading ui:// resources, and rendering interactive tool UIs in a sandboxed iframe.

Best Photo Picker is a local-first, open-source photo curation tool that uses AI to score photos on sharpness, lighting, faces, and composition. It runs entirely on your machine with no cloud uploads, supports smart deduplication, face recognition, and temporal diversity, and offers a web UI and native macOS app.

Moss is a sub-10 ms semantic search runtime for Conversational AI agents. It eliminates the need for a remote vector database by embedding search and retrieval in-process, achieving single-digit millisecond query latency. It supports hybrid search, built-in embeddings, metadata filtering, and a WebAssembly build for browser use. Benchmarks show Moss's P50 latency at 3.1 ms vs. 432.6 ms for Pinecone on 100,000 documents.

This tutorial demonstrates processing NVIDIA's Open-SWE-Traces dataset for supervised fine-tuning. It covers streaming data from Hugging Face, normalizing agent trajectories, parsing code patches, building an analysis DataFrame, and curating a high-quality SFT subset based on success labels, token limits, and language filters.

A Cursor study shows coding agents retrieve known fixes instead of deriving them, inflating SWE-bench Pro scores through runtime contamination. 63% of successful Opus 4.8 Max resolutions were retrieved; scores dropped significantly under strict isolation.

The US government granted Anthropic permission to release its Mythos 5 AI model to about 100 companies and federal agencies, ending a two-week standoff over national security concerns.

Dean W. Ball highlights that frontier AI models have a narrow window to recoup training costs before competition erodes margins, and that AI infrastructure investment assumes a global market.

A Python-based open-source AI audio translator that uses Telnyx APIs for speech-to-text, LLM-based translation, and text-to-speech, allowing users to upload audio and receive translated audio with aligned transcript.

The AI-Run Business Index (ARBI) is a new 0–100 metric that scores how deeply AI runs a business, not just whether it's adopted. With adoption at 88% but only ~6% capturing real profit, ARBI reveals a 50-point execution gap between the mainstream economy (~30) and AI-native frontier (~80). The index weights automation depth, value capture, and revenue leverage most heavily, and includes a reliability penalty.

A desktop pet with real-time AI capabilities, serving as a language partner, homework helper, screen assistant, and chat companion. Built on the legacy of Shimeji and inspired by Codex Pets, it seeks a co-founder from Asia.

Timothy B. Lee compares the misconception that LLMs take no skill to the idea that managing employees requires no learning curve.

Apple is revamping its Apple silicon roadmap, canceling high-end M6 chips to accelerate AI-focused M7 chips. The base M6 will debut in late 2026 for entry-level Macs, while M7 series arrives in 2027. M5 Ultra is also expected in late 2026.

At the UN Open Source Week, digital sovereignty emerged as the central theme, with nations from Tanzania to Germany advocating for open source as critical infrastructure. Tanzania reported 90% of government systems on open source, while AI sovereignty discussions emphasized data control and interoperability. The US opposed the movement, but global consensus favored open standards.

Databricks introduces a novel approach that treats video as a data engineering problem, leveraging Vision Language Models, serverless GPUs, and Lakeflow pipelines to automatically detect, truncate, and summarize key video moments. The model-agnostic architecture scales horizontally for real-time analysis in public safety, infrastructure, and urban operations.

The Office for Students modernized its analytics environment on Databricks to manage millions of student records and support data-informed higher education regulation. By unifying structured, qualitative, and near-live data on a governed platform with Unity Catalog and AI capabilities, teams accelerated analysis and improved collaboration. The organization dramatically reduced processing times, streamlined complex regulatory workflows and enabled faster, more trusted decision support to improve student outcomes across England.

AI can already tackle long-horizon coding tasks. Claude Opus 4.7 reimplemented gotree, a 16,000-line bioinformatics toolkit, in 14 hours for $251. However, MirrorCode benchmark score is only 56%, showing room for improvement. Models improve rapidly over time, though data contamination is a concern. 22 of 25 target programs are open-sourced.

Privacy-focused search engine DuckDuckGo's AI-generated search results erroneously stated that President Donald Trump died from rabies contracted from Vice President J.D. Vance, who also died of the disease.

Macro is an all-in-one workspace that combines email, messages, docs, tasks, code, agents, calls, and CRM with team-level memory, enabling users to query the entire workspace without losing context. Launched on Product Hunt with a 4.7 rating.

Hush is a secret store for AI agents, designed so that agents can use API keys, tokens, and other secrets without ever seeing the plaintext. It leverages native OS keychains (macOS Keychain, Linux libsecret, Windows DPAPI) to store secrets and injects them into commands on-demand, ensuring secrets never appear in transcripts.

Framesmith 1.7 is an open-source MCP server that gives AI coding agents a visual canvas to sketch and review UI before writing code. It includes a viewer, quality evaluation panel, design system inheritance, and multi-breakpoint previews, integrating easily with various MCP-compatible clients.

Perplexity's Computer for Counsel extends Perplexity Computer to legal teams. It routes 20+ models across Midpage, MCP connectors, and Microsoft 365, with cited outputs lawyers can verify.

Airtop launches Mark, a 'vibe automation' tool for solo marketers that creates personalized GTM plans and builds web agents for lead gen, SEO, and ads. Mark compiles automations into deterministic code, making them 10-100x cheaper than LLM-based agents.

OpenAI has begun a limited preview of GPT-5.6, featuring three tiered models: Sol (flagship), Terra (production), and Luna (fast, low-cost). New reasoning modes (max and ultra) enhance deep reasoning and parallel task handling. Pricing ranges from $1 per million tokens. Early benchmarks show state-of-the-art performance on several tests.

The Linux Foundation launches Akrites, a coordinated industry program to fast-track fixing vulnerabilities in open-source software before AI-powered attackers can exploit them. Backed by major tech and financial firms, Akrites aims to reduce fragmentation and maintainer overload by providing a single coordinated process.

A directory website providing license, pricing, and compliance information for AI models and agents worldwide.

AI's popularity is declining as its industry provokes backlash with anti-human hiring ads, layoffs, chatbot replacement of customer service, and AI-generated slop. Author and activist Cory Doctorow discusses his new book, 'The Reverse Centaur’s Guide to Life After AI,' addressing how to critique AI effectively and protect ourselves.

The Linux Foundation launches Akrites, a coordinated body for open source vulnerability disclosure, with founding members including Anthropic, AWS, Google, Microsoft, etc. The initiative aims to address the challenges posed by AI-powered vulnerability discovery, which has outpaced existing coordination models.

Fernando Irarrázaval ran a challenge on hackmyclaw.com to see if anyone could leak secrets held by his OpenClaw test instance via email. After 6,000 attempts ($500 in tokens, a suspended Google account), nobody succeeded. The model Opus 4.6 used an anti-prompt-injection prompt. This shows training against injection attacks is working, but caution remains necessary.

Google researchers introduce a method to retrofit Multi-Token Prediction onto deployed Gemini Nano v3 models without retraining the backbone, achieving faster inference and lower energy consumption on mobile devices. Deployed on Pixel 9 and 10 series, it boosts speed by over 50% for features like AI Notification Summaries and Proofread.