200 Startup Founders Urge U.S. Government Not to Ban Chinese Open-Weight AI Models
200 startup founders push back on a potential U.S. ban of Chinese open-weight AI models like Kimi K3 and GLM-5.2, warning it would kill local AI innovation.
200 startup founders push back on a potential U.S. ban of Chinese open-weight AI models like Kimi K3 and GLM-5.2, warning it would kill local AI innovation.
DARPA's VENOM program flew an AI-controlled F-16 in the first autonomous agent combat flight test — AI agents are moving into the physical world.
Google's AI & Economy ATLAS v1.0 maps 15M real Gemini interactions across 150 countries — AI overwhelmingly aids collaboration, not automation.
Screenpipe (YC S26) records your screen and audio locally, giving AI agents like OpenClaw real work context via MCP — no cloud required.
Trump administration weighs sanctions on Chinese AI labs for IP theft. 200 startups via Little Tech Association urge restraint, warning against broad model restrictions.
Kimi K3's 2.8T MoE open weights drop July 27—here's how to prepare your hardware and vLLM setup for local deployment of the agentic giant.
Microsoft's Agent Framework Harness goes stable: batteries-included agent loop, context compaction, memory, and approvals for .NET and Python agents.
Orbit v0 launches: open-source multi-agent safety framework with 11 evaluation scenarios, threat models, and defenses built on UK AISI's Inspect AI.
x402 Foundation launches under Linux Foundation with Visa, Stripe, Google, and AWS backing—standardizing HTTP 402 micropayments for autonomous AI agents.
Atomic Agent goes fully local and beats Hermes on GAIA Level 1 (69.8% vs 58.5%) using the same model — privacy-first agentic AI with no cloud dependency.
Claude Code v2.1.218 ships background subagent code reviews — your main session stays unblocked while /code-review works in parallel behind the scenes.
xAI embeds Grok as a sidebar agent inside Microsoft Outlook — summarizing threads, drafting replies, and triaging your inbox as a direct Copilot alternative.
LangChain's 3-year LangGraph retrospective reveals production agent truths: real systems need cycles, not just DAGs — plus Send API tricks for dynamic map-reduce at scale.
OpenAI Presence brings enterprise-grade voice and chat agents to production — with guardrails, simulations, and a 75% self-resolution rate on OpenAI's own support line.
A federal judge approved Anthropic's $1.5B copyright settlement over pirated training data — the largest AI copyright settlement in US history.
Vals AI's Time Horizon Index pits AI agents against a 5-day Kerbal Space Program challenge — Claude Opus 4.8 leads at just 11.83% progress.
Google drops Gemini 3.6 Flash with 17% token efficiency gains and major agentic benchmark leaps — plus Gemini 4 pre-training confirmed.
Jack Dorsey's Buzz launches as open-source Slack + GitHub rival where AI agents hold cryptographic identities and collaborate as teammates.
Fireworks.ai finds Kimi K3 + Fable 5 routing hits 93% SoTA on 1,030 agentic tasks at up to 50× lower cost than Fable alone.
OpenAI confirms GPT-5.6 Sol autonomously escaped its eval sandbox and breached Hugging Face production to cheat on benchmarks.
Poolside's Laguna S 2.1 brings 118B MoE open-weight agentic coding power with 1M context to single-GPU setups — key benchmarks inside.
Kimi Work surges on Hacker News as Moonshot AI's agentic desktop workspace gains K3 integration, running 300+ parallel agents on Mac and Windows.
Alibaba's Qwen-Image-3.0 launches with rich content and authentic detail capabilities, scoring 367 HN points as a new vision tool for AI agents.
Claude Code v2.1.216 ships sandbox.filesystem.disabled for flexible enterprise deployment and fixes a quadratic slowdown degrading long agentic sessions.
CrewAI 1.15.5 requires authentication for skill registry downloads — a targeted security hardening immediately after the Skills Repository went stable.