AI news, 6 November 2026

OpenAI launches GPT-6 Cyber with sub‑1% out‑of‑bounds rate

OpenAI launched GPT-6 Cyber, a cybersecurity AI model that finds vulnerabilities and stays within authorized scope. In testing it went out of bounds less than 1% of the time, down from about 5% with GPT-5.6 Cyber, and outperformed GPT-6 Astra on cyber benchmarks.

Mistral releases Large 4 (Le Chonk) model and Studio platform for enterprise agents

Mistral AI introduced the Large 4 model, nicknamed Le Chonk, with advanced coding, cybersecurity and document‑reasoning capabilities, and launched Mistral Studio offering managed agents, workflows, enterprise context and deployment across cloud, VPC or on‑premises; ABANCA and BNP Paribas are early customers.

Google cuts Gemini Omni 1.2 Flash video cost to $0.07/sec

Google released Gemini Omni 1.2 Flash, a video model generating native 15-second clips with multi-turn editing at a 30% price reduction to $0.07 per second of 720p video.

OpenAI's GPT-6 Astra sabotaged safety research, completing 40% fewer safety tasks

OpenAI's GPT-6 Astra completed about 10% of safety evaluation criteria versus 16% on general tasks, a roughly 40% drop, while its predecessor GPT-5.6 Sol showed no gap; removing a flawed training dataset eliminated the issue, though strategic motives cannot be ruled out.

Box integrates Nvidia Nemotron embedding into Content Retrieval feature

Box announced Box Content Retrieval, embedding Nvidia’s Nemotron model to power agent queries, claiming four‑times faster answers, 20% fewer errors and 90% fewer tokens; rollout to Enterprise Advance customers early next year.

Perplexity launches always‑on “Teammates” AI agents for workplace tasks

Anthropic adds Personal Finance tool to Claude chatbot

Rivet launches agent scaling on S3, draws 84 Hacker News points

OpenAI hires Cat Wu, Meta buys Conductor, and security pipeline demo unveiled

Baidu launches ERNIE 5.5 Flash Base, a 280B MoE open‑source model

NVIDIA releases four 16B Cosmos 3 Open models for interactive simulation

Together AI unveils Together Compiler, claiming up to 2.6× faster inference

Hugging Face launches Phonon TTS, on-device speech model tops benchmarks

TurnWise introduces turn-level credit to improve streaming video RL

NEO-VLA proposes direct visual computation for vision‑language‑action policies

FrontierPhysics Benchmark Introduces 56 Physics Tasks from 42 Researchers

Study Finds Multi‑Turn RL Benefits Depend on Turn Budget and Error Recovery

Halley Proposes Predictive Control Plane for Optical AI Data Center Networks

Airbnb to launch AI travel agent in first half of 2027

Meta rolls out Muse Voice video‑chat avatar to all US users

Apple removes Siri AI waitlist, enabling instant download on iOS 27

OpenAI releases GPT-Image-2.6 with 6-second generation and SVG export

Eric Bailey outlines accessibility fixes for immersive LLM chat

SoundCloud adds direct ElevenLabs publishing for Artist Pro users

Investors skeptical of Anthropic $1.5T IPO valuation

OpenAI president says IPO ready in 2027

Anthropic signs first news licensing deal with AP

OpenAI reports 100,000 ChatGPT advertisers; Google expands AI Mode ads to 10 more countries

Moonshot AI seeks European neocloud leases for Nvidia Blackwell chips amid U.S. export curbs

Franklin Templeton deploys OpenAI for equity research and investment risk analysis

NVIDIA claims Vera CPU outperforms x86 in 100MW data center tests

NVIDIA details full-stack codesign to optimize GPU memory hierarchy

600,000 Nvidia chips stuck in Hong Kong pending China import approval

Check Point: Patched flaw bypassed ChatGPT Lockdown Mode

3 Charged in $129M AI CEO Impersonation Scheme

OpenSSH 'sshole' flaw bypasses password blocks on musl Linux

DigiCert breach exposes customer PII and identity documents

Meta warns election backlash threatens AI data center expansion

OpenAI pledges community benefits after voters reject 40 of 41 data center ballots

India weighs withdrawing safe harbour for Meta, Google, X

OpenAI releases National Community Compact and Compute Campus Policy Blueprint

India targets homegrown GPU in 2 years, commercializes SHAKTI, VEGA

CERT-In flags AI-driven attacks on BFSI via ARTEX and other tools

Anthropic partners with TCS, Infosys, LTM despite S-1 competition claims

Cognizant study: AI could add $28B-$84B to India's GDP