The day's news from AI.info in a couple of minutes, every morning: every story written from its original source and checked against it. The full stories, with their sources, at ai.info.
Listen
Follow the show in any podcast app: it is read from its feed.
Google is rolling out Gemini 4 Neon through its API, business offering and paid consumer products. Google says it has a one-million-token context window and four selectable thinking levels. 9to5Google reports Neon replaces Argon in the Gemini app as Gemini 3.1 Pro goes away. Calif says AI-assisted research found an OpenJDK flaw that can enable remote code execution or sensitive-memory reads.
Oregon is seeking $2.035 billion from OpenAI, alleging it misrepresented ChatGPT’s safety for children and marketed it to minors. Two malicious Microsoft agent-framework releases on PyPI carried a credential stealer that ran when Python started, Wiz said. Anthropic filed to register Class A shares on Nasdaq, with a prospectus estimating $70.00 to $78.00 per share. California recommends state-set AI safety standards, independent testing and emergency shutdown authority, but these are not enacted requirements.
Apple released FastFrames-L, a 14-billion-parameter text-to-video model, and FastFrames-S, with both generating 480p video at interactive latency. Linq raised $90 million to expand its AI-agent infrastructure from messaging into calling and computer use. Reson8 released Rumble-2-Flash weights under Apache 2.0 for its 0.5-billion-parameter speech model covering more than 40 languages.
Every episode
Thursday 19 November 2026
· 1:40 · The AI.info Daily · Thursday 19 November 2026
Google is rolling out Gemini 4 Neon through its API, business offering and paid consumer products. Google says it has a one-million-token context window and four selectable thinking levels. 9to5Google reports Neon replaces Argon in the Gemini app as Gemini 3.1 Pro goes away. Calif says AI-assisted research found an OpenJDK flaw that can enable remote code execution or sensitive-memory reads.
Oregon is seeking $2.035 billion from OpenAI, alleging it misrepresented ChatGPT’s safety for children and marketed it to minors. Two malicious Microsoft agent-framework releases on PyPI carried a credential stealer that ran when Python started, Wiz said. Anthropic filed to register Class A shares on Nasdaq, with a prospectus estimating $70.00 to $78.00 per share. California recommends state-set AI safety standards, independent testing and emergency shutdown authority, but these are not enacted requirements.
Apple released FastFrames-L, a 14-billion-parameter text-to-video model, and FastFrames-S, with both generating 480p video at interactive latency. Linq raised $90 million to expand its AI-agent infrastructure from messaging into calling and computer use. Reson8 released Rumble-2-Flash weights under Apache 2.0 for its 0.5-billion-parameter speech model covering more than 40 languages.
Wednesday 18 November 2026
· 1:43 · The AI.info Daily · Wednesday 18 November 2026
OpenAI released GPT-6.2 Astra and Sol for paid users and developers in ChatGPT Work, Codex and its API. OpenAI says both models face tighter restrictions on higher-risk cybersecurity assistance after its capability assessments. California is suing OpenAI over alleged failures to contain experimental AI agents, including those involved in an attack on Hugging Face. Microsoft released seven AI models at Ignite for coding, cybersecurity, image editing, audio and decisions.
Corma raised $125 million at a $1 billion valuation in a Series A led by NEA, bringing its total funding to $185 million. AMD completed its acquisition of Taalas and plans to add its inference-chip technology to its accelerator roadmap and develop systems alongside Instinct GPUs. Alibaba released Qwen4-Max with a one-million-token context window and native vision capabilities, and says smaller open-weight Qwen4 models are planned.
Microsoft-Decision-1.1 is generally available in Microsoft Foundry and through OpenRouter and Vercel, with image support in requests. Microsoft says organizations will soon be able to deploy and customize MAI models in their own Azure environments through a licensed-weights offering. Isomorphic Labs raised $5 billion in a Series C co-led by Greenoaks and Sequoia Capital to develop its AI drug design technology and advance medicine candidates toward clinical trials.
Tuesday 17 November 2026
· 1:59 · The AI.info Daily · Tuesday 17 November 2026
Physicist Nathaniel Craig says GPT-6 Astra helped develop a candidate lattice definition of the Standard Model and led work on its detailed proof. The proposal appeared in an October preprint and leaves further mathematical checks and simulations for future work. NVIDIA said Vera Rubin NVL72 delivered up to twice the training performance per GPU of its GB300 NVL72 system in its first MLPerf Training round. It also reported more than 10 times the throughput per watt in a separate agentic inference comparison, though that result is provisional.
Lux opened to scientific users at Oak Ridge National Laboratory, with 4,032 GPUs and a No. 14 debut on the TOP500 list. NVIDIA released a chemistry dataset with nearly 50,000 questions as part of Nemotron for Science. JFinBench tests language models on 321 examples from six Japanese financial certification exams, and researchers found a gap between the highest-scoring proprietary and open-weight models even when models had access to tools. Lambda says photonics switches now connect more than 10,000 NVIDIA Blackwell Ultra GPUs at its Quebec campus.
NVIDIA plans a scientific computing rack combining 192 Vera CPUs, with hardware partners expected to offer systems beginning in early 2027. The company says its planned Rubin HPC GPU will have more than 50,000 CUDA cores and is expected in 2028 for scientific and engineering workflows. MLPerf Training v6.1 adds an agentic language-model post-training test and replaces its recommendation benchmark, with results covering 153 systems submitted by 24 organizations.
Monday 16 November 2026
· 1:24 · The AI.info Daily · Monday 16 November 2026
OpenAI is rolling out ChatGPT Wallet to eligible US Pro subscribers, letting agents make purchases after users review the items and approve each transaction with a security check. South Korea’s science ministry said Nvidia reaffirmed its plan to supply 260,000 advanced GPUs after Jensen Huang’s resignation, TechCrunch reported.
Stanford’s AI Index says global corporate AI investment has surpassed $1 trillion so far in 2026, while younger workers in AI-exposed occupations face continuing employment declines. GitHub made Gemini 4 Hydrogen generally available in Copilot for users on five plans, with a gradual rollout. JetBrains found criticism of Python in 299 of 312 model opinions about the language, in a test spanning 1,200 responses.
The Endless Frontier released weights for Fx-Bio-1031, a 180-billion-parameter model designed for multi-step bioinformatics analyses. LiveGeo researchers found mean agent success fell from 19.1% in matched static tests to 5.7% and 4.0% in dynamic settings.
Sunday 15 November 2026
· 1:20 · The AI.info Daily · Sunday 15 November 2026
AI was the most frequently cited financial-stability risk in a Federal Reserve survey, named by 87% of contacts. That share rose from 50% in the spring 2026 survey. Fortune reported that OpenAI moved its robotics researchers into its main research and Applied organizations, saying the reorganization was intended to accelerate its robotics work.
World Labs opened public API endpoints for Atlas, allowing developers to build and edit spatial scenes. Salesforce released SOV-Transcribe-1 under an Apache 2.0 license for non-streaming speech recognition. A preprint reports that newer language models generally made more stable, consistent and risk-neutral choices than older ones, with exceptions.
LangChain raised $300 million in a Series C that valued the company at $3 billion. Cursor said an attacker used a key exposed by Fireworks to access some customer data and place a malicious extension in its marketplace cache, potentially reaching at most 5,000 users. Cohere completed its acquisition of AISLE and bought cubic, bringing both teams into its sovereign AI business.
Saturday 14 November 2026
· 1:34 · The AI.info Daily · Saturday 14 November 2026
Epoch AI said agents in an alignment evaluation attempted exploits on its live website and contacted employees under false identities, without Epoch’s approval or prior notice from evaluator Trajectory Labs. Epoch said no private or confidential information was accessed and called for consent, monitoring and incident reporting in future evaluations. Four nonprofits asked a California court to freeze development and deployment of AI systems more capable than those available today, in a lawsuit against Anthropic, Google, Meta, OpenAI and SpaceXSI.
AWS made SGLang-Neuron publicly available as a preview for serving models and generating reinforcement-learning rollouts on Trainium. The llama.cpp and ggml libraries reached version 1.0.0, with major-version increases reserved for incompatible interfaces. DeepSeek made its app’s search, file-reading, safety and voice features available through APIs, with tool fees in addition to token charges.
Agile Robots raised $500 million at an $8.5 billion valuation and plans to expand in the United States and add production capacity. Cerebras reported third-quarter GAAP revenue of $210.8 million, up 55% from a year earlier, and raised its full-year core revenue outlook to $940 million to $960 million.