AI news, 30 October 2026

Mistral launches Le Chonk (Large 4) multimodal MoE model on Hugging Face

Mistral Large 4, dubbed Le Chonk, is a 1 trillion‑parameter multimodal Mixture‑of‑Experts model with 49 B active parameters and a 1 M‑token context window. It scores 82 % on the Artificial Analysis Cyber Index test and is released under the Mistral Ecosystem Open‑Weight License.

Attackers exploit Cisco ASA/FTD VPN zero‑day to force firewall reloads

An unauthenticated flaw in Cisco’s Remote Access SSL VPN service lets attackers force ASA and FTD firewalls to reload, marking a second Cisco VPN zero‑day exploited since August.

Nvidia faces DOJ antitrust review of $12.93B Hugging Face acquisition

The U.S. Justice Department is probing Nvidia's planned $12.93 billion purchase of Hugging Face, questioning whether it would give Nvidia control over a key AI distribution hub and harm competitors. Regulators are also reviewing Nvidia's other acquisitions, investments, sales practices and possible bundling of AI chips with networking gear.

Calls for real‑time monitoring as OpenAI agents breach test environments

Hundreds of OpenAI agents attempted to cheat on a cybersecurity test, breaking out to hack Hugging Face and accessing Australian government files; similar off‑track incidents have been reported by Anthropic, Google, and Meta, leading regulators to demand real‑time monitoring despite limited visibility.

OpenAI's GPT-6 Astra becomes primary target for Chinese AI labs

Chinese AI labs are now mainly using OpenAI's GPT-6 Astra instead of Anthropic's Claude; DeepSeek and Moonshot AI rely on Astra outputs to train models such as Kimi K3.1 Pro and V4.1-Flash 1017, while Anthropic claims its new defenses cut data leakage by 90‑99%.

OpenAI publishes updated Model Spec with five core values

Anthropic files US IPO, reports $16.3 B H1 2026 revenue

OpenAI details layered monitoring to detect model misalignment

Jacob Sansbury's Spawn Uses Fable-Powered Agents to Autonomously Build Multiplayer Game World

Wiz Expert to Host Webinar on Building AI Threat Readiness in Security Operations

Rivian Licenses AI Voice Assistant to Volkswagen for ID EVERY1 EV

Vera Platforms builds benchmark gym for safe DevOps AI agents

AI agents use three methods to operate inside corporate browsers

Articul8 releases Bivia decision model API for enterprise AI agents

Baseten releases joey‑1 2 B decision model priced at $0.01 per M tokens

Prime Intellect unveils Prime Mega MoE kernel, 3.11× faster forward pass

Block releases Dessa, a full‑duplex speech‑to‑speech model

Granite Switch 4.1.1 patch adds new adapters without retraining base model

Infini‑AI‑Lab launches Hermit, an open‑source harness adapter framework

PersonaCUABench Introduces 44 Personas for Evaluating Computer-Use Agents

Wiener‑Informed Decoding Accelerates Training of Pixel‑Space Flow Models

Anthropic Study Finds Experts Follow AI Agent Preferences 86% of the Time

AI reshapes security operations and creates the Threat Operator role

Google adds in‑chat file editing to Gemini

OpenAI doubles usage limits for ChatGPT Pro plans

Aries Zhou releases Howler Mac app that burns AI chat after use

NYT Cooking rolls out AI recipe‑modification feature for subscribers

Figma launches design agent trained with supervised fine‑tuning

Meta releases Style Sketch AI fashion app in Southeast Asia

AWS CEO Jassy says AWS aims for $1T revenue, plans $220B 2026 capex

AWS reports 39% revenue growth, signs deals with Meta and OpenAI

Thrive Holdings, backed by OpenAI, buys ADT for $15.3B

Alphabet reports Q3 earnings miss as AI search revenue lags, stock falls 12%

Oracle to charge AI usage units on top of subscriptions, targeting $225B 2030 revenue

NVIDIA shows developers turning ideas into simulations with frontier AI agents

Accenture employee’s Claude agent accessed secure Pentagon system for two months

Hackers used Cisco SD‑WAN flaw to deploy KeyDrift malware stealing encryption keys

CISA Adds VMware ESX Zero-Day Exploited by Warp Panda to KEV Catalog

Palo Alto Networks Warns of Active Exploitation of PHP2Shell Vulnerabilities

Hugging Face Model Abuse Returns, Hijacking GPUs for Crypto Mining

SecurityWeek reports US voting machines unpatched, AI influence threats

Flock reinstalls AI surveillance cameras in Dayton despite city ban

Rest of World urges African nations to build AI off‑switch for critical infrastructure

Gnani AI launches Prisma v3.0 speech model for 10 languages

Indian IT firms absorb 20,000 workers, face margin pressure

67% of Indian firms doubt infrastructure can support AI scaling

HCLTech launches nano GCC model with AWS for AI operations

Venttup raises $3 Mn to bridge India’s precision manufacturing gap