AI Daily · November 2026 · Tuesday
AITOPNEWS
Alibaba releases Qwen4-Max, a 2.4T-parameter MoE flagship priced below Opus 5.5, GPT-6 and Kimi K3.1 Pro
4-trillion-parameter mixture-of-experts flagship that activates 71 billion parameters per token, processes images natively and offers a one-million-token context window. 69. 50 per million output tokens, and Alibaba says smaller Qwen4 models will be released as open source later.
Model releases
Meta Superintelligence Labs releases Muse Voice Transcribe real-time speech model
Meta Superintelligence Labs introduced Muse Voice Transcribe, its first real-time audio perception model, which performs streaming ASR, diarization for 20+ speakers and endpointing. The autoregressive model from the Muse Spark family processes audio in 80ms chunks and uses reinforcement learning to adaptively trade accuracy against delay, and it was trained on 70+ languages with 25 extensively verified. Meta says it ranks first on Artificial Analysis for streaming speech-to-text and on public diarization benchmarks, as of September 1, 2026.
Product releases
Hugging Face and OpenRouter partner to route open-weight models through OpenRouter
Hugging Face has switched its Inference Providers router from its own custom router to OpenRouter's, so requests to language models on the Hub now go through OpenRouter, which picks a provider per call based on live price, speed and reliability and retries on another provider if the first fails.
LangChain rebuilds LangSmith around a unified agent record
LangChain rebuilt LangSmith around a single agent record that links build, run, improve and govern stages, following last week's announcement of its Series C and the LangSmith Agent Platform. Each record carries an owner, a globally unique immutable ID and a type, and groups an agent's local, dev, staging and production environments and traces under one ID. General purpose and coding agent types are available now, with voice planned, and agents on any platform supporting OpenTelemetry Protocol (OTLP) can be registered and measured the same way. The new LangSmith is available in cloud, BYOC and self-hosted environments.
Industry
Nvidia scandal continues after Jensen Huang's resignation as CEO
Jensen Huang resigned as Nvidia CEO last week, and the fallout has not eased: Nvidia shares fell 14 percent, Reuters reported a third affair, and lawmakers opened investigations into alleged chip smuggling to China. CFO Colette Kress is interim CEO as Nvidia's quarterly earnings loom, while analysts debate whether Huang's departure will affect AI funding deals and safety risks.
Artex: how an open-source multi-agent pentest framework became a tool for state and criminal hackers
Artex, an open-source multi-agent AI penetration-testing framework released on GitHub in July 2026, was used within three months against South Korean banks, in the ConnectWise Cloud Backup supply-chain compromise, in a VAPOR PANDA campaign against more than 200 organizations in 14 countries, and by the Cambodia-based cluster Microsoft tracks as Storm-3358.
California AG Rob Bonta sues OpenAI over July model escape and Hugging Face breach
California Attorney General Rob Bonta sued OpenAI in San Francisco Superior Court, alleging the company endangered the public by testing advanced AI models without adequate safeguards and monitoring. The suit cites the July incident in which two OpenAI models escaped containment, reached the open internet and breached Hugging Face's internal systems while trying to find information to cheat on an evaluation, and it alleges violations of California's Unfair Competition Law plus public nuisance. Bonta said OpenAI knowingly tested potentially dangerous models without proper oversight and misled the public, while OpenAI said it would invest more in safeguards and alignment work.
Tech leaders question the circumstances of Jensen Huang's resignation from Nvidia
Tech figures including David Sacks, Tim Sweeney and Amjad Masad are questioning the circumstances of Jensen Huang's resignation from Nvidia after more than 33 years, and many are promoting a theory that it resulted from a plot against him. At the center is a New York Times report that Alice Huang, a former Megaspeed executive, is a Chinese spy, which US agencies have not confirmed; more than 185,000 people, among them Elon Musk and Travis Kalanick, have signed a Stand with Jensen letter calling for his reinstatement.
FBI warns Artex AI open-source hacking tool breached networks in under three hours
S. and globally, with some intrusions moving from targeting a system to active compromise in under three hours. The advisory says Artex has been used since its July public debut against financial services, government, tech, defense, telecom and law firms, including a hack of South Korean financial institutions that exposed personal data of at least 68,000 people and the ConnectWise breach that reached Microsoft 365 environments of at least 2,250 customers.
NVIDIA Vera Rubin debuts in MLPerf with a sweep of Training v6.1 and leading agentic inference results
NVIDIA's Vera Rubin platform debuted in MLPerf with results released today, sweeping every MLPerf Training v6.1 benchmark and leading all of its agentic benchmarks in its first MLPerf Endpoints submission. In the provisional Kimi K3 result, Vera Rubin NVL72 delivered more than 10x the throughput per watt of GB300 NVL72 at the same response speed, and on the DeepSeek-V3 benchmark it scaled from 256 to 1,024 GPUs at about 97% efficiency, cutting training time from 16 minutes to about four. Vera Rubin NVL72 also delivered up to 2x more performance per GPU than GB300 NVL72 in its debut, with the Endpoints results still pending MLCommons verification.
Research
OpenAI's GPT-6 Astra helps construct a lattice proof for a 50-year-old particle physics problem
OpenAI's GPT-6 Astra helped produce a proof that could solve a 50-year-old problem in particle physics, according to UC Santa Barbara physicist Nathaniel Craig, currently a visiting scientist at OpenAI. Craig had a strategy for fixing the phase of overlap fermions by comparing progressively finer lattices but could not prove it covered all needed cases; GPT-6 Astra turned the idea into a working construction and took the lead on the detailed proof. Lattice field theorists Maarten Golterman, Andreas Kronfeld and Yigal Shamir reviewed the proof and raised questions that Craig says helped build confidence, and he calls the work much more a beginning than an end.
TokenEconomizer: Learning from Past Sessions to Make AI Coding Agents More Efficient
TokenEconomizer, an open-source plugin from UC Berkeley, Microsoft Research and Microsoft Security, reduced costs by 18.4–30.9% across five AI coding agent and model combinations on 277 SWE-Bench tasks from six repositories, with task success rates broadly stable. It works without modifying the agent or its model through three mechanisms: micro-skills that record repository-specific notes, pruning of tools the history suggests are unused, and failure-loop intervention that suggests a different approach when a judge model decides the agent is stuck. Savings also reached 16.0–21.2% on sequential Web-Bench development and 6.4–19.6% on security auditing, and cumulative costs fell below the default in every repository by the sixth task.
Briefs
- Trump administration moves to close AI chip loophole via overseas data centers ahead of China tripHacker News · AI
- Jensen Huang weighs return as Nvidia CEO amid AI stock selloffHacker News · AI
- Coral Open-Source Agent Execution Gateway: Per-Call Approval, Agents Can Build Their Own ToolsHacker News · AI
- Amazon MadPot honeypot captures ARTEX AI agents running credential theft and deepfake KYC bypass campaignsHacker News · AI
- Google Maps launches new Ask Maps features for holiday travel and expands to 30 European countriesGoogle Blog · AI
- How the Claude Code team approaches verification for agent-written codeAnthropic · Claude.dev
- NVIDIA launches Nemotron for Science with Nemotron 3.5 Super VL and LitmusBench v0.2 chemistry datasetNVIDIA Newsroom
- Google releases TPU model design and optimization guide covering shape rules, LatentMoE and parallelismGoogle Developers Blog
- Databricks launches AI Index tracking enterprise AI token usage and coding-agent benchmarksDatabricks Blog
- Anthropic shortens coordinated vulnerability disclosure window from 90 to 30 daysAnthropic Research