AI News
The desk's complete reporting record, newest first.
Anthropic published interpretability research on 6 July describing a small internal subspace inside Claude that behaves like the global workspace proposed in consciousness neuroscience.
Meituan disclosed LongCat-2.0 on 2 July, a 1.6 trillion parameter mixture of experts model, and attached one detail that will be studied in Washington more closely than any benchmark.
Together AI closed an $800 million Series C on 1 July at a valuation of $8.3 billion, led by Aramco Ventures with participation from NVIDIA, Vista Equity Partners and General Catalyst.
Anthropic released Claude Sonnet 5 on 30 June, describing it as its most agentic Sonnet to date and pricing it to make the flagship optional. The model plans, drives browsers and terminals and runs autonomously at a level that until recently required larger and more expensive systems.
The Council of the European Union gave its final approval to the Digital Omnibus on AI on 29 June, completing a legislative process that began with a Commission proposal in November 2025. The European Parliament had adopted the text on 16 June.
OpenAI unveiled Jalapeno on 25 June, its first custom inference accelerator, designed with Broadcom. The company claims nine months from design start to tape-out, a reduction of roughly 50 per cent in inference cost, and a planned deployment scale of 1.3 gigawatts.
Four senior researchers left Google DeepMind inside a single week in June, and the DeepMind talent exodus has become the most closely watched personnel story in the sector. Noam Shazeer, co-inventor of the Transformer architecture and a Gemini co-lead, departed for OpenAI on 18 June.
Two releases on 18 June changed the arithmetic for anyone paying for a frontier coding model. Z.ai published GLM-5.2, a 753 billion parameter mixture of experts model under an MIT licence with a one million token context window.
Anthropic says Claude subscriptions will no longer cover third-party tools such as OpenClaw. The immediate commercial shift is clear. The strategic read is harder to ignore.
Reports that DeepSeek V4 will run entirely on Huawei chips point to more than supply-chain improvisation. They suggest China’s AI stack is becoming more vertically coherent under pressure.
Gemma 4 is Google’s latest push into the open-model market, and the company is pitching it as its most intelligent open family to date.
The Netflix VOID model has arrived on Hugging Face with a clear video editing use case: video object and interaction deletion. Here is what the Netflix VOID model does and why the release matters.
Solana Agent Skills give AI builders a cleaner way to work with Solana programmes, tokens and developer tooling. Here is what Solana Agent Skills does, why it matters, and what developers should watch next.
Chinese startup Agibot staged a live robot-led show featuring over 200 humanoids from its A2, X2, G2, and D1 series performing autonomously under broadcast conditions.
Andrej Karpathy open-sources Autoresearch, a 630-line Python framework that lets AI agents autonomously run and iterate on machine learning experiments overnight on a single GPU.
Bltz AI, founded by former CrowdStrike leaders, launches a self-healing agentic AI security platform with runtime protection against prompt injection, data leakage, and misuse.
Microsoft releases practical guidance for security teams on detecting prompt abuse, investigating unexpected AI behaviour, and applying enterprise safeguards against injection attacks.
NVIDIA launches NemoClaw at GTC 2026, an open-source enterprise AI agent platform that adds security and privacy to OpenClaw, installable in a single command alongside OpenShell security runtime.
OpenAI releases GPT-5.4 mini and nano, designed for high-volume coding assistants, subagents, and computer use at near-flagship performance and much lower cost.
Xiaomi officially launches the MiMo-V2 family, revealing MiMo-V2-Pro as the mystery Hunter Alpha model on OpenRouter, a trillion-parameter reasoning model for agentic workloads.
Perplexity brings its AI-powered Comet browser to iOS, making its built-in Comet Assistant available to iPhone users for the first time.
Telecommunications companies are leading the charge in agentic AI adoption, with 48% of telco respondents reporting active deployment of autonomous AI agents, according to NVIDIA’s 2026 State of AI in Telecommunications report.
A court temporarily reversed an injunction blocking Perplexity’s Comet shopping agents from Amazon, in a test case for agentic commerce and AI agent access to retail platforms.
LangChain partners with NVIDIA to deliver a production-grade agentic AI development platform, combining LangSmith, Deep Agents, and NVIDIA Agent Toolkit for enterprise-scale deployment.
SoftBank and OpenAI have officially launched their long-anticipated joint venture, SB OAI Japan GK, headquartered in Tokyo.
End of the AI archive · Open the full Newswire · RSS
