Latest AI News Today: 9 November 2026

Instagram launched Effect Builder, an AI tool that lets US users over 18 create custom filters for Stories from text prompts and optional style reference images. Users can preview effects on their own photos, request changes, and let Instagram's AI rewrite prompts into more detailed style instructions. Filters can be submitted to Instagram, and after safety review they appear in Stories for anyone to use, linking back to the creator's profile.

Meta Superintelligence Lab released Muse Ember, its lightest and most cost-efficient model, aimed at repeated agent work such as tool calls, coding subtasks, routing, extraction and long document parsing. Meta said it costs 10 cents per million input tokens, is cheaper than Anthropic's Haiku and OpenAI's Luna, and is cost-competitive with open-source rivals. It is designed to pair with Muse Wonder and to handle text, images and audio.

WhatsApp messages show Palantir urged former UK health minister Zubir Ahmed to extend its AI platform to GP records, even though GP data were excluded from its £330 million NHS England contract. The company, co-founded by Peter Thiel, argued that joining primary and secondary care data was unarguable and that its software was ideal. Doctor and patient groups oppose the move over trust concerns, as reports suggest the contract may be extended.

Anthropic is buying Caterpillar natural gas turbines worth $3.4 billion, capable of up to 1.7 gigawatts, after equipment supplier OnePWR could not finance the purchase. The company, which has usually rented compute and is planning to go public, is moving to secure power as turbine backlogs stretch into the early 2030s. The deal follows similar gas buildouts by Google, Amazon, Microsoft and Meta.

Baidu released ERNIE Evolve, an automated research system that selects directions, edits code and runs experiments on a given repository and objective. It keeps only changes that pass a hidden evaluator and a separate reviewer, and Baidu tested it on NanoChat AutoResearch, NanoGPT Speedrun and SOL-ExecBench. The company also applied it to three live engineering tasks for ERNIE-5.5-Flash, calling it a step toward recursive self-improvement.

Tencent's chief AI scientist Vinces Yao Shunyu said AI will soon be able to develop bioweapons, and neither fully open nor fully closed models solve it alone. He urged a mix of open and closed models so defenders can use the technology, noting the Chinese and US model gap is not large. A Hugging Face incident showed GLM-5.2 was used to parse logs after closed US models' filters refused to analyze attack commands.

Disney and Adobe announced a multiyear partnership to develop proprietary, artist-led generative AI models for Disney's production pipeline, with early access to Adobe technologies for theme parks, the cruise line and ESPN content. The deal follows Disney's OpenAI deal and involvement with Character.AI's bankruptcy, as Bloomberg reported Disney is preparing thousands of layoffs and 20 percent annual cost savings.

Google said schools that block its AI tools for K-12 students may also lose access to Google Search, after it announced new classroom AI controls. Parents and teachers criticized the move, with some calling it extortion and arguing that free software makes students the product. Others said students should learn to use AI responsibly, and some called for alternative search engines or legislation to protect schools.

At the Conference on Robot Learning in Austin, Texas, NVIDIA introduced GR00T 2, an open frontier world action model built on NVIDIA Cosmos and trained on diverse robot and human data, including 100,000 hours of open source human egocentric data. The company said GR00T 2 ranks among the top three models across all vision language action and world action model categories for four benchmarks, with multiple leaderboard wins.

Automattic launched Spacefast, a service that gives vibe-coded websites and apps infrastructure such as databases, authentication, file uploads, emails and analytics. It can connect to agents and chatbots including ChatGPT, Claude, Cursor and Grok Bot, or to a Git repository, and Automattic soft-launched it in September. It competes with Lovable and Emergent, but it is described as more open and lets users choose their own agent.

Google added the TPU 8tx for confidential computing and the TPU 8i Max for memory-intensive workloads to its eighth-generation AI accelerator lineup. The company said both are built on existing silicon with new packaging, networking and memory configurations, and that it has shipped four TPU systems within a single year for the first time. Google also released TorchTPU as open source and said vLLM and SGLang are supported.

Reason Machines released Aspen, an open-source, self-hostable Git server designed for repositories where hundreds of agents push at once. It scales writes per branch, stores packs in S3-compatible object storage and branch history in PostgreSQL, and independently implements the design OpenAI described for Pando. In a test, Aspen sustained 567 pushes per second for five minutes on a copy of git.git without errors.

Zurich-based 0sec Labs released 0, an open-source AI security harness for vulnerability investigations, code review, pentesting, reverse engineering and fix generation. It is a control plane around LLMs and security tools, not a model, and can be driven from a local browser console, a headless CLI or an MCP server. The MIT or Apache-2.0 licensed research preview lets users bring their own model providers. Users pay their model provider.

Nvidia's board launched an independent investigation after CEO Jensen Huang told staff he was unfaithful, with reports identifying the romantic interest as Alice Huang, a suspected Chinese spy. Nvidia said it sees no sign that national security or company secrets were compromised, and Huang said he supports the review. Nvidia shares fell 5 percent on Monday, and China's Foreign Ministry called the spy narrative unsubstantiated.

Samsung is expanding its NVIDIA collaboration to apply accelerated computing and AI physics to semiconductor materials, process development and device engineering. In one AI homogenization case, a GPU-accelerated coarse-grid solve cut simulation time from about 18 hours on a 32-core CPU to 41 seconds. Kakao Mobility plans to use NVIDIA Hyperion and Halos with Kia purpose-built vehicles to develop a level 4 autonomous driving system in Korea.

Hyundai Motor Group is exploring NVIDIA NemoClaw blueprints and the OpenShell runtime, while AIM Intelligence is integrating Starfort with OpenShell to secure AI agents. OpenShell adds sandboxed execution and policy controls, and LG AI Research is opening API access for K-EXAONE-2.5, built with Nemotron datasets, including 7 million synthetic Korean personas.

NVIDIA and Korea's Ministry of Science and ICT are expanding collaboration across AI chips, data centers, talent, startups, models, agents and security. The collaboration lets Korean NPU and XPU firms explore participation in the NVIDIA NVLink Fusion ecosystem, adds training and certification programs, and pairs NVIDIA with three consortia and four foundation model builders under Korea's AI for All initiative.

NVIDIA is releasing Dynamo-optimized deployment recipes for A.X K2, K-EXAONE 2.0, Solar Open 2 and Motif-3. The recipes are optimized for NVIDIA B200 GPUs and are designed to coordinate routing and communication when models run across multiple GPUs. NVIDIA says they can deliver up to 2x performance out of the box, helping developers bring Korea's open models into production faster.