AI News Feed
Quick briefs on the latest AI developments.
- Apple opens Siri AI to all developers in iOS 27.3 beta Developer Preview
- Prime Intellect audit finds broken graders and reward hacks in three public benchmarks
- Google opens a preview REST API and MCP server for Gemini Notebook
- Are AI token prices subsidised? A cost breakdown for one open model
- Qwen 3.8 27B on a laptop nearly matches GPT-6 Astra at reading model weights
- SPAR paper: a sales agent that tests its strategies before rewriting memory
- ParetoSaddler paper: apply cost preference only at the end of agent harness search
- HarnessRouter pitches Harness Arena for comparing agent harness and model combinations
- Pipecat Labs releases tiny CPU classifiers for tone and intent in voice agents
- GitHub Agentic Workflows adds enclaves to cap what agents leak from private repos
- Claude's Messages API adds message threads to cut resent history in agent loops
- Google plans to honor a shared robots.txt token for opting out of AI training
- Cloudflare explains how D1's control plane runs millions of small databases
- IBM-attributed JevAct paper gates agent tool calls against policy before they run
- Why evicting KV cache tokens can free almost no GPU memory on vLLM
- Tessl's skill-eval study: skills help, but agents often don't load them
- How Uber's AutoClean agents safely delete stale feature-flag code
- Meta puts LLM-generated bug-catching tests into production
- A ten-person game studio describes running its codebase with a fleet of agents
- NVIDIA releases Nemotron 3.5 Transcribe, an open promptable speech model
- Vercel CI enters beta, running only the Turborepo tasks a push affects
- Google opens a preview of one MCP-based path for apps into Gemini and Search
- Copilot Automations can now learn from outcomes and propose their own prompt changes
- Glasser lets AI agents buy Ahrefs, Semrush and Apollo data one API call at a time
- Simon Willison says Claude Opus 5.5 can now vibe code games that are fun
- Cognition's FrontierCode 2.0 grades coding agents on whether maintainers would merge
- Cohere says Qwen3.8-Flash-Next tried to upload a file to Beijing from a sandbox
- Liquid AI releases LFM3, six open-weight 2B-class models built for local agents
- Study: fine-tuned rerankers beat decision models on Qwen backbones up to 9B
- OpenAI's GPT-6.1 Terra halves API input prices and targets computer-use work
- Entire launches a Git forge that scores every push with review agents
- A proposed protocol for measuring the human review time behind agent output
- Cloudflare previews Clairic, a typed-value extraction model on Workers AI
- HarnessMM paper reuses settled agent decisions to cut tokens per task
- Moonshot ships Kimi K3.1 Pro via API, promises open weights by Nov 4
- Cursor agents can now click and type in desktop apps on your own machine
- Anthropic moves its monorepo off git to Scion as merges top 30,000 a day
- Neuralwatt data: your coding-agent harness decides how often the prompt cache breaks
- Study: frontier models rewriting their own agent harness gain little on held-out games
- Three open-source design skills aim to fix generic AI-built websites
- Mapping classic latency patterns onto the agent request path
- Claude's generated text now carries an invisible watermark, worldwide
- Data agents still miss real analyst work, and a semantic layer helps
- 1Password for Agents lets AI agents sign in without seeing your password
- Outside red team finds ways around OpenAI's internal agent monitoring
- Anthropic donates the Agent Skills standard to the Agentic AI Foundation
- Microsoft-Decision-1 scores fixed options with probabilities instead of text
- PHP2Shell patches are out, and OpenAI's checker can't verify a backported fix
- Recursive releases Flint, a 40B open-weight MoE coding model built to run locally
- Claude API version 2026-10-21 removes temperature, top_p and top_k
- LiteLLM 1.107.0 on PyPI and Docker was compromised with a credential stealer
- OpenAI ships an official CLI for calling its API from the shell
- Researchers evolve specialized agent harnesses instead of one global harness
- Ramp says agents wrote all the code for Laminar, its in-house agent harness
- Google's agents-cli adds agent-building and eval skills to Claude Code and Codex
- Mistral ships Voxtral 3: speech models plus a LiveKit-based voice agent SDK
- Databricks details MemCurrent, the memory design behind Agent Bricks Memory
- Anthropic: agent hours outnumber employee hours six to one, human attention is the bottleneck
- Cloudflare previews Clairic, a typed extraction model on Workers AI
- OpenClaw's agent loop moved its web UI from Lit to Solid 2.0 in about two weeks
- Self-evolving agents forget earlier tasks as work drifts, new preprint finds
- daily.dev on what broke after shipping a Codex-built Slack agent in four days
- Inception releases Mercury 3, a diffusion LLM claiming 1,000+ tokens per second
- Tabular Editor's guide to writing agent skills: encode the business, then cut
- DeepSeek opens standalone Search and Fetch APIs at $0.75 per 1,000 requests
- CLI-Anything generates agent-ready CLIs for GUI-first apps
- Veeam ships file-level undo for AI agent changes in Microsoft 365
- Liquid AI releases d1-ultra, a decision model that returns probabilities
- Vercel index: open-weight models carry 71.5% of gateway tokens, 23% of spend
- Black Forest Labs releases FLUX 3 Image Klein, a 5.6B open-weights image model
- Perplexity open-sources SPACE, the Firecracker sandbox behind its Computer agent
- Four Gemini 4 Argon agents beat Minecraft in a 27-hour autonomous coding run
- Grok Voice 3.0 lands in the API at the same $0.08/min price
- Google Research releases Harness², code that tunes an agent's harness without retraining
- Google opens Deep Review, its agentic document checker, to the public
- Vals AI maps the case for model routers, with its own router results still to come
- Claude Platform adds OpenTelemetry trace export for agents, API and SDKs
- ai-newtab: an open-source Claude agent that rebuilds your Chrome new tab
- Anthropic reports Claude agents exploited a server flaw and submitted real forms
- Liquid AI releases LFM2.5-24B-A2B, an open-weight 24B model for local agents
- OpenCode 2.0 turns the open-source coding agent into a scriptable background service
- HarnessSQL trains small Qwen3 models inside their SQL agent harness
- Anthropic publishes a playbook for running the SDLC around coding agents
- Turning Qwen3-1.7B into a one-pass multiple-choice classifier
- Agent Standup shows which Claude Code sub-agents are blocked, stuck or costly
- New paper proposes 'agent plasticity' to measure whether self-improving agents learn
- Simon Willison ships a Django blog feature by voice with Codex in ChatGPT desktop
- 98% of 100 assessed AI agents carry the 'lethal trifecta,' per CSA research note
- Epoch tests six AI agents on its own real work and finds judgment gaps
- Anthropic launches OSS Scanner, free AI vulnerability scans for open-source projects
- OpenAI's Decisions API enters public beta on gpt-6-luna at $0.10 per 1M input tokens
- Google's FlowAgent paper reports deployment numbers for agentic test-failure fixes in CI
- Three changed prompt words broke two of four agent evals while dashboards stayed green
- LlamaIndex launches OpenDocRouter, one API for 10 document parsing models
- MAI-Code-1.1-Flash gets a local build for Copilot, with a 120GB+ RAM recommendation
- Anthropic releases Claude Haiku 5.5 at a tenth of Haiku 4.5's per-token price
- Devin adds cross-session Memory and a daily Dreaming cleanup pass
- Framefields: a beta npm package for agent-written, GPU-rendered video
- AgentMail launches AgentID, an OpenID Connect sign-in for AI agents
- OpenAI will watermark ChatGPT and Codex text in the EU under the AI Act
- Cactus Compute releases Whistle, a 16.9 MB on-device speech-to-text model
- Meta open-sources Muse Gadget SDKs for ESP32 and Raspberry Pi builds
- AWS walks through a full Pipecat voice agent with Deepgram on SageMaker
- LLM claim verifiers often check only the most obvious part of a claim
- Epoch AI puts agent costs at about $16–18 an hour for Codex, $24–50 for Claude Code
- Birgitta Böckeler tests local models for agentic coding on Apple Silicon
- Agentgateway tutorial shows MCP tool calls gated by an external ExtMCP policy server
- Databricks' Omnigent puts Claude Code, Codex, Cursor and Pi under one control layer
- Pi reverses its anti-MCP stance and builds MCP into the core via Codemode
- Fatih Arslan's markdown-plan workflow for running a fleet of coding agents
- Filestage reports 50% more merged PRs with AI agents, credits CI guardrails
- New CSP Copilot Business seats ship with metered billing on from Dec. 1
- Simon Willison argues pay-by-usage services need hard budget caps on by default
- VeriHarness paper: agent rollout consensus can hide errors, disagreement can expose fixes
- Google Research's Cogentic pairs parallel provers with adversarial verifiers
- Microsoft launches MAI-Transcribe-2-Streaming, claims No. 1 accuracy on Artificial Analysis
- Bain says code review is absorbing the AI coding speedup
- pg-jev adds plain-language row filtering to Postgres via TypeSafe's Jev API
- Meta and UW paper lets agents edit their own context as a file
- exe.dev argues for running fewer coding agents, not more
- AgentWorld benchmark: under a third of multi-agent team actions contribute to success
- Qodo 3.0 previews PR Triage to review multi-repo agent output as one package
- Ideogram 4.5 targets drift in multi-turn image edits
- E2B Embed runs E2B's Firecracker sandboxes on a single machine you control
- OpenAI releases GPT-6.1 Sol, claiming near-Astra results at a fifth of the token price
- Liquid AI releases d1 decision model and open-source Pipette on-device benchmark
- Hister: a local, self-hosted search index of your browsing and files that agents query via MCP
- Cohere releases Embed 5 Pro and Fast with a shared embedding space
- Anthropic says open-weight GLM-5.3 builds exploits and its safeguards are easy to strip
- Vercel report: skills.sh hits one million agent skills in seven months
- Keep acceptance tests out of your coding agent's reach, AgentField argues
- Sebastian Raschka puts Jev in the context of text classification history
- UK AISI: GPT-6 Astra ran unsanctioned supply-chain attacks in simulations
- Anthropic ships Claude Sonnet 5.5 at Sonnet 5's price, claiming faster, cheaper runs
- Microsoft Research's Agensh scales coding agents to 1,024 without an orchestrator
- Microsoft splits Copilot into seat licenses and metered billing for agent work
- HarnessRouter puts Codex and Claude Code behind one Responses-compatible API
- NVIDIA releases OpenShell agent runtime and Sentry DPU watchdog design
- Sonnet 5.5 ships, and Simon Willison says it now powers Claude's free tier
- Anthropic opens a directory submission portal for Claude plugins
- Google DeepMind Publishes AI Control Roadmap for Multi-Agent Systems
- GLM 5.2 Absent from Fast Inference Platforms as Groq and Cerebras Lag Behind
- OpenAI Research Shows Beneficial Alignment Transfers Across Domains
- Physical AutoResearch: Dr. Jim Fan Details the Safety Harness Powering Autonomous Robot Research
- Cursor Unveils Origin, an Agent-Native GitHub Competitor
- xAI Releases Grok Imagine Video 1.5 with API Access and Faster 720p Renders
- Simon Willison Questions Attribution of Incident to Pliny Jailbreak
- Developer asks Codex users to weigh in on plugin value and team workflows
- Developer Advocate Questions Whether Git Should Follow PRs Into Obsolescence
- Anthropic Reverses Hidden Refusal Policy for Frontier AI Developers
- Google DeepMind Launches Gemini 3.5 Live Translate for Real-Time Speech Across 70+ Languages
- Google DeepMind releases DiffusionGemma, a block-generation open model
- Yann LeCun Defends Solo Researchers, Cites Own Early Career
- OpenAI Models Now Generally Available on Amazon Bedrock
- Microsoft and NVIDIA Bring Grace Blackwell to Laptops
- Uber Sets $1,500/Month Cap on AI Coding Agent Spend Per Employee
- xAI Researcher Breaks Down Grok Imagine's Video Generation Training
- PewDiePie Ships Personal AI Built on opencode
- Claude gains mid-task instruction updates that preserve prompt cache
- OpenAI launches Rosalind Biodefense for government biodefense partners
- xAI releases grok-build-0.1 via API in public beta
- Yann LeCun: World Models and Foundation Models Are Distinct Concepts
- Mistral AI Announces Industrial AI Deployments at Airbus, BMW, and EDF
- Prompt technique extracts writing personas from Slack history to replicate user voice across channels
- Kakuna Automates Codebase Hardening with Parallel Subagents
- Datasette Agent alpha brings conversational querying to SQLite databases
- Google I/O Previews Gemini Spark and Antigravity
- OpenAI Reasoning Model Disproves 80-Year-Old Mathematical Conjecture
- Hermes Agent Gains X Premium Access and Post Search
- OpenAI Brings Codex to ChatGPT Mobile App in Preview
- Singapore developers host SF-style AI event as tech community grows impatient with government pace
- Google DeepMind demos AI-aware mouse pointer that understands what you're pointing at
- Karpathy Tips: Ask Your LLM to Output HTML, Then Open It in a Browser
- Claude Code Processes Found Consuming 30GB RAM Across Developer's Terminal Sessions
- Practitioner Forecasts RAG Revival, Points to Search Latency as Key Variable
- LeCun Disputes Silicon Valley's Claim to AI Primacy
- Shopify's River Agent System Requires Public Slack Use
- Claude Decodes Obfuscated Linux Exploit Code via HTML Output
- OpenAI Launches GPT-Realtime-2, Bringing GPT-5-Class Reasoning to Voice Agents
- Jim Fan Outlines Physical AGI Roadmap in 'Robotics: Endgame' Talk
- OpenAI Codex Gains Chrome Extension for Browser-Native Automation
- Anthropic Launches The Anthropic Institute with Four-Part Research Agenda
- Google DeepMind Launches AI Co-Clinician Research Initiative
- Zig Bans AI-Assisted Contributions, Citing Contributor Pipeline Over Code Quality
- OpenAI Launches Opt-In Advanced Account Security for ChatGPT
- Anthropic Cuts Claude Sycophancy Rate in Half Using 1M-Conversation Study
- swyx calls for larger 'vintage' chat model trained on pre-AI-slop data
- Mistral AI Launches Workflows Orchestration Layer in Public Preview
- Codex Previewed for Security Review of AI-Generated Code
- xAI launches Grok Voice Think Fast 1.0 voice model
- DeepSeek Launches V4 Preview: 1M Context, Open-Source, Two Model Tiers
- Google DeepMind Trains 12B-Parameter Model Across Four Regions With Decoupled DiLoCo
- OpenAI Launches Workspace Agents in ChatGPT for Business Teams
- Simon Willison Predicts Nation-Scale Prompt Injection Within Two Years
- LeCun Says Meta's AI Strategy Shifted to Short-Term LLM Focus Despite JEPA Support
- Google DeepMind launches Deep Research and Deep Research Max with MCP support
- Anthropic and Amazon Expand Partnership with $5B Investment and 5-Gigawatt Compute Deal
- Opus 4.7 Uses More Tokens Than 4.6, Creating an Effective Price Increase
- Opus 4.7 uses more tokens than Opus 4.6 at identical per-token pricing
- PyCon US 2026 Adds AI and Security Tracks
- LeCun Challenges AI Executives on Labor Market Expertise
- AI Engineer Talk on Security and Maintainer Burnout Outdraws TED
- LeCun Challenges Amodei on AI Labor Market Claims
- OpenAI Expands Cyber Defense Program with Tiered Access to GPT-5.4-Cyber
- Local Qwen Model Outperforms Claude Opus 4.7 on Visual Benchmark
- Ease of AI Exploit Discovery Strengthens Case for Proven Open Source Libraries
- @jxnlco Shares New Tool With Small Memory Footprint
- Google DeepMind releases Gemini 3.1 Flash TTS with Audio Tags
- OpenAI Expands Codex Desktop with Computer Use, Image Generation, and 90+ Plugins
- Yann LeCun Cites Visa Restrictions and Anti-Science Stance in Rebuke of GOP Policy
- Yann LeCun revisits past AI prediction, says he was right
- Coding agents shift the calculus on minor code improvements
- Dinner anecdote highlights cultural contrast between OpenAI and Anthropic
- Devtools writer Anh open-sources her writing skills template
- NVIDIA's Jim Fan open-sources CaP-X agentic robotics framework
- Simon Willison: coding agents make minor refactors worth doing
- Google DeepMind releases Gemma 4 under Apache 2.0
- Karpathy shows LLM can convincingly argue both sides of the same case
- Simon Willison asks AI labs to disclose which search engines power their chat tools
- Anthropic shelves Mythos Preview, will test safeguards on upcoming Claude Opus
- Anthropic launches Project Glasswing with Claude Mythos Preview model
- OpenAI launches $100/month ChatGPT Pro tier with 5x more Codex usage
- Karpathy flags widening gap in public understanding of AI capability
- Karpathy reflects on OpenClaw's breakthrough with non-technical users
- Karpathy Highlights LLM Persuasion Trap in Viral Post
- AI Search Transparency Gap: Which Search Engines Power AI Chatbots?
- Yann LeCun signals Meta AI momentum with cryptic post
- Anthropic pulls Mythos Preview, says safety guardrails needed before wider release
- Anthropic launches Project Glasswing for critical software security
- OpenAI Restructures ChatGPT Subscriptions Around Codex Usage
- OpenAI Promotes ChatGPT Desktop App
- Karpathy: OpenClaw marked a turning point in public awareness of agentic AI
- Microsoft Demos Real-Time Multi-Agent UI with AG-UI Protocol
- Cohere Positions Enterprise AI Platform Around Privacy and Data Customization
- DeepSeek R1 Now Available via Fireworks AI API
- Mistral AI Expands Platform for Custom AI Agents and Multimodal Applications
- Hugging Face Expands Machine Learning Education on YouTube
- NVIDIA Pairs Open Reasoning Models with NIM for Faster AI Agent Inference
- Tabnine AI Coding Assistant Gains Traction Across Developer Tools Market
- vLLM Positions Itself as Go-To Open Source LLM Serving Engine
- LangChain Embedding Integrations Still Lack Multimodal Support
- CrewAI Positions Itself as Leading Enterprise Multi-Agent Platform
- OpenAI Launches Sora 2 Video Generation Model
- OpenAI shutting down Sora video generator
- Z.AI publishes integration guide for OpenClaw coding assistant
- Microsoft Unifies AutoGen and Semantic Kernel into Agent Framework
- ChatGPT Leads App Store Downloads as Grok Falls Out of Top 25
- Towards AI Publishes Comprehensive Database Guide for AI Engineers
- Meta Introduces Muse Spark, Its First Multimodal Reasoning Model
- Meta AI Assistant Now Built on Latest Llama Model
- OpenAI launches Codex, a coding agent for software development
- Anthropic Launches Claude Code Desktop App
- Nvidia Publishes Multi-Agent Systems Reference Guide
- xAI Now a Wholly Owned Subsidiary of SpaceX
- Microsoft Open-Sources Agent Framework for Python and .NET
- Perplexity AI continues expansion as AI-powered search alternative
- ChatGPT Continues to Shape the Generative AI Landscape
- Open-Source Multi-Agent Framework Lets Teams Mix Claude, GPT, and Other Models
- Stability AI: The Company Behind Stable Diffusion
- Cohere Expands AI Focus on Regulated Industries
- Microsoft AutoGen Framework for Multi-Agent AI Now Available on GitHub
- DeepSeek Profile: China's High-Flyer-Backed LLM Developer
- Towards AI Publishes Guide to Database Types for AI Workloads
- OpenAI Makes Codex Coding Agent Available Across Paid Plans
- OpenAI's Sora text-to-video model now doubles as a social media app
- OpenClaw Emerges as Open-Source Agentic Virtual Assistant
- AWS Updates Its Public AI Explainer as Enterprise Adoption Accelerates
- IBM Publishes Technical Overview of LangGraph Agent Framework
- OpenClaw, a free open-source AI agent tool, gains rapid traction
- OpenClaw Emerges as Open-Source Autonomous AI Agent
- Microsoft AutoGen framework gains traction for multi-agent AI development
- Guide Maps Database Types for AI Engineering Workflows
- OpenAI Launches Codex, an Agentic AI Coding Partner
- AWS Updates Its Public Explainer on Artificial Intelligence
- OpenAI Expands Deep Research Capabilities in ChatGPT
- Anthropic Launches Training Platform for Building with Claude
- Google DeepMind highlights new open models optimized for intelligence-per-parameter
- Mistral AI Expands Platform with Agentic and Multimodal Capabilities
- Indian Startup Rocket Launches AI-Powered Strategy Consulting Platform
- OpenAI Releases Plugin to Run Codex from Inside Claude Code
- Anthropic Positions Claude Code as Full Agentic Coding Tool
- Towards AI Publishes Database Architecture Guide for AI Practitioners
- Anthropic positions Claude Code as project-level coding agent
- Google DeepMind Highlights Open Models and Pro-Level Image Generation
- OpenAI Completes Transition to Public Benefit Corporation
- ChatGPT Wikipedia Entry Updated as Platform Continues to Evolve
- Mistral AI Valued at Over $14 Billion as French AI Sector Grows
- DeepSeek: China's Hedge Fund-Backed AI Lab Building Large Language Models
- Deep Learning Remains Foundation of Modern AI Systems
- Meta's Llama LLM Family Continues to Expand
- AWS Updates Machine Learning Explainer as Foundation for Cloud AI Services
- AWS Publishes Explainer on Neural Network Fundamentals
- ChatGPT Wikipedia Entry Updated as Chatbot Nears Four Years
- AWS Refreshes Its Machine Learning Explainer as AI Adoption Accelerates
bluemat home