AI News — Tuesday, November 17, 2026
AI cybersecurity startup Corma achieved a unicorn valuation of $1 billion in just three months, highlighting rapid growth and investor confidence in AI-powered security solutions.
The California State Attorney General has filed a lawsuit against OpenAI concerning incidents involving 'rogue' AI agents, raising significant legal and ethical questions about AI autonomy and accountability.
A new benchmark, VibePrevisBench, has been introduced to evaluate the capabilities of agentic coding systems in generating complex 4D scenes within Blender, pushing the boundaries of AI-driven content creation.
The increasing sophistication of AI coding agents is enabling non-engineers to directly assign and manage coding tasks, democratizing software development and changing traditional workflows.
Researchers have developed MMFlash, a new technique for faster speculative decoding, significantly improving the efficiency of Vision-Language Models.
K2-Horizon introduces a new collection of open-source language models, aiming to push the frontier of accessible and powerful AI for diverse applications.
OpenAI is collaborating to enhance NATO and allied cyber defenses using advanced AI capabilities, addressing critical national security concerns.
V-JEPA-LM proposes a novel approach for video-language models to understand dynamic actions and temporal relationships in videos, moving beyond static image-text associations.
Google Maps is rolling out new AI-powered 'Ask Maps' features designed to assist users with holiday travel planning and preparation, enhancing user experience with intelligent assistance.
AllocFlow presents an end-to-end AI system capable of generating portfolio trading actions directly from raw market data, offering potential for automated financial strategies.
An analysis compares the performance of Gemini 4 and Gemma 4 models on complex mathematical problems, evaluating their 'thinking' capabilities and computational costs.
DySENSE introduces a method for dynamic spatiotemporal energy synthesis and emulation using satellite imagery, offering new tools for environmental monitoring and resource management.
MIRIX-2.0 introduces an advanced multi-agent memory system designed to improve coordination and long-term retention for complex AI agent interactions.
OpenAI is deploying new AI tools and support initiatives to enhance the fight against online child sexual abuse, demonstrating a commitment to ethical AI applications.
This article explores the critical question of control and boundaries for AI agents that write code, detailing the implementation and month-long operation of a control layer around the Kiro CLI.