1. Google launches Gemini 3.8 Flash Cyber, opens automated vulnerability finding and patching to more than 650 government and enterprise partners Google has launched Fairwind, a restricted-access program that gives more than 650 government agencies, companies and cybersecurity partners worldwide access to Gemini 3.8 Flash Cyber.

2. Enterprise’s self-hosted small model takes over half of internal AI traffic, team says it handles 116 million requests a month An enterprise AI team says it has consolidated requests from more than 200 internal applications onto one smaller, self-hosted language model.

3. ZimaBlue Trains Robots on 120,000 Hours of First-Person Video, Lifting Zero-Shot Manipulation Success From 36.1% to 77.8% A research team says ZimaBlue, a framework for training “world action models,” can turn large volumes of first-person video into robot-control capabilities.

In Brief

• Trump Administration Backs OpenAI’s Fair-Use Argument in Times Lawsuit The administration filed a statement of interest supporting OpenAI’s argument that training language models on copyrighted text can qualify as fair use. The New York Times alleges that OpenAI and Microsoft unlawfully used its articles and seeks billions of dollars in damages.

• OpenAI Faces 30 New Lawsuits Over Tumbler Ridge Shooting Students and educators filed 30 federal lawsuits accusing OpenAI and CEO Sam Altman of substantially assisting the alleged shooter by failing to act on flagged ChatGPT conversations and prevent renewed access. OpenAI strategy chief Jason Kwon called claims about the company’s safety decisions false.

• New York City Restricts Classroom AI Through Eighth Grade New York City announced a one-year moratorium covering roughly 600,000 public-school students in 2-K through eighth grade beginning in the 2026–2027 school year. The rules also prohibit AI grading, ban companion chatbots across all grades, and limit high-school AI use to approved cases and pilots.

• Palo Alto Networks Reportedly Paid $500 Million for IT-Agent Startup Console Palo Alto Networks acquired Console, whose AI agents automate help-desk work such as password resets and troubleshooting, for $500 million in cash and stock, according to TechCrunch sources. The cybersecurity company plans to integrate Console into its Cortex platform for natural-language alert investigation and resolution.

• Wonderful Raises $550 Million at a $5 Billion Valuation Israeli-Dutch startup Wonderful more than doubled its valuation in under six months with a Series C led by Insight Partners and joined by Salesforce. Its Wonderful AI OS connects enterprise agents and workflows with company data and existing systems, and the funding will support product development and deployment teams.

• HiddenLayer Secures $100 Million for AI Runtime Security HiddenLayer raised a $100 million Series B led by Delta-v Capital after reporting that annual recurring revenue grew more than tenfold over the past year. Its tools protect models, agents, and workflows against threats including prompt injection, agent manipulation, malicious tool use, and compromised open-source model files.

• Amazon Adds AI-Based Verification for Messages Claiming to Be From Amazon Alexa for Shopping can now compare an email, text, or phone call with Amazon’s messaging records while checking its sender, formatting, and contents. Amazon says the assistant confirms authenticity only when completely certain and otherwise directs customers to their orders and official support.

• Reliance Jio Opens Its Cloud PC Service Across India Reliance Jio made JioPC available to Indian internet users regardless of their broadband provider, streaming a virtual computer with configurations of up to eight virtual CPUs, 16GB of RAM, and 1TB of storage. Jio says the subscription can give computers as old as eight years access to cloud capacity for modern AI applications.

• OpenAI Endorses California Youth AI-Safety Bill OpenAI announced its support for California SB 1119, describing the bill as establishing age-appropriate AI safeguards for teenagers while preserving opportunities to learn, create, and explore.

• Google Reportedly Seeks AI-Training Licenses From Major Hollywood Studios Google has approached studios including Disney, Warner Bros. Discovery, and Universal about licensing copyrighted libraries for AI training, according to unnamed sources cited by the Los Angeles Times; no agreements have been announced. The reported proposals could include character-specific payments and shares of advertising revenue from AI-generated content.

• Qwen-Drive Unifies Perception and Planning for Autonomous Vehicles Researchers introduced Qwen-Drive-1.0, a vision-language foundation model that combines 3D perception, driving-scene question answering, and future-trajectory generation while retaining the underlying pretrained architecture. They report competitive motion-planning results across open-loop, pseudo-closed-loop, and closed-loop evaluations.

• UI-Venus-2 Targets Automation Across Mobile, Web, and Desktop UI-Venus-2 is an open-source multimodal agent built around a unified reasoning-and-action loop, with coverage spanning more than 170 multilingual mobile apps and native desktop systems. Its training framework uses trace- and sample-level verification plus safety mechanisms for consequential actions.

Read the full edition →