August 28, 2026 —
Federal judge overturns the Pentagon’s Anthropic blacklist
Today’s North America AI brief is led by a consequential legal ruling on how the U.S. government may treat an AI supplier that resists unrestricted military use. The rest of the issue shows the industry pushing outward at the same time: agents are gaining a common interface to physical equipment, security leaders are coordinating around AI-enabled cyber risk, and new products are moving generative AI deeper into video, travel, education and scientific work.
Federal judge overturns the Pentagon’s Anthropic blacklist
U.S. District Judge Rita Lin ruled that the Pentagon acted unlawfully when it labeled Anthropic a supply-chain risk and imposed broad penalties after the company challenged unrestricted military use of Claude. The 59-page decision found unlawful retaliation and procedural failures, and it also addressed orders directing other federal agencies to stop using Anthropic products. The government is expected to appeal.
The ruling does not require the military to buy Anthropic’s technology, but it limits the government’s ability to use procurement and national-security authorities as punishment for a supplier’s public policy position.
Anthropic proposes a common control layer for physical AI
Anthropic opened a research preview of the Model Hardware Standard, a model-agnostic specification that lets AI agents discover, understand and operate programmable equipment through standardized drivers. Early partners are testing it across microscopes, liquid handlers, robotic arms and quantum-computing hardware. Anthropic says integrations that once took weeks or months can be reduced to hours or minutes, while safety limits and device characteristics remain explicit in each driver.
AI and security leaders call for a coordinated cyber-defense surge
More than 100 organizations, including OpenAI, Anthropic, Google, Microsoft, AWS, major security vendors and financial institutions, signed a joint call for action on AI-enabled cyber risk. The group urged operators to fix high-risk weaknesses, expand access to cyber-capable AI for defenders, share verified fixes and threat intelligence, and give critical-infrastructure teams stronger support during major incidents. The letter warns that the useful lead defenders hold may last only months.
Google makes Gemini Omni 1.1 Flash production-ready
Google released Gemini Omni 1.1 Flash as the generally available version of its conversational video generation and editing model. The update can extend an existing clip, interpolate between first and last frames, and generate at 360p, 720p, 1080p or upscaled 4K. It accepts up to ten seconds of video context for editing or extension, and Google is exposing the stable model through the Gemini API for production workflows.
Anthropic opens 10,000 discounted Claude seats for scientists
Anthropic is opening 10,000 Claude team-plan seats for academic and nonprofit research labs for one year. Standard access will be free, while premium seats with five times the usage limit will cost $15 per month. The company also widened its AI for Science grants beyond biology, allowing eligible projects to apply for as much as $50,000 in credits, while keeping its most capable life-science models behind additional access controls.
Google’s AI Mode moves from travel answers toward transactions
Google added conversational flight-price tracking, points-and-miles pricing and hotel booking to AI Mode in Search. Users can monitor fares from more than 300 airline and travel partners, compare rewards options and move from hotel discovery into room selection and checkout with integrated providers. Flight tracking is available in more than 180 countries, while the booking experience begins with selected partners and keeps the travel provider responsible for the transaction.
Khanmigo adds Gemini-built interactive diagrams and teacher controls
Google and Khan Academy moved two Gemini-powered classroom features into use: responsive math and science diagrams for students, plus editable AI-generated practice materials that teachers review before assignment.
OpenAI and Thailand launch an AI startup accelerator
OpenAI and Thailand’s higher-education ministry started an eight-week accelerator for ten health, wellness and education startups, combining API credits, technical mentoring, evaluation work and a November demo day.
Microsoft offers Copilot access to one million Filipino teachers
Microsoft and the Philippines Department of Education expanded their partnership to give one million public-school teachers Copilot access for lesson planning, personalized learning, administrative work and classroom support.
Aolani launches pay-per-token inference in Southeast Asia
Singapore neocloud Aolani introduced Token Factory, a managed inference service with OpenAI-compatible APIs, elastic capacity and usage-based billing for open-model families including DeepSeek, GLM, Kimi and Qwen.
GeForce NOW adds user controls for DLSS 4.5
NVIDIA previewed new DLSS 4.5 controls for GeForce NOW Ultimate, along with Firefox support, broader device coverage, GOG single sign-on and five additional launch-day game commitments.
OpenAI details how applied AI engineers bridge models and customers
An OpenAI Forum panel described Applied AI Engineering as a customer-facing discipline that combines product expertise, real-world testing and field feedback to turn frontier models into dependable deployed systems.