Google Selects 10 Startups for Africa AI Lab Cohort

Key Takeaways

• The FAA’s nationwide SMART system analyzes 200 data streams; the agency says it helped one Tampa-to-Dallas flight save 42 minutes, while lawmakers raise testing and oversight concerns.
• Google selected 10 African startups for its first Applied AI Lab cohort, offering access to Gemini and Gemma, mentorship and investor support ahead of a December demo day in Accra.
• Keir Lieber says nuclear arms-control precedents have limited relevance to AI, given private-sector incentives and the absence of a world government; a midterm report describes AI regulation as bipartisan after the 2026 elections.
• The United States and China plan AI safety talks during Donald Trump’s meeting with Xi Jinping in Shenzhen on November 18–19, including model testing, human oversight and a channel for serious-incident reports.
• MCP and A2A connect AI agents but do not replace enterprise identity, permissions, secrets management, trusted context, workflow redesign or monitoring.
• Businesses adopting AI in 2027 will need to manage data access and decision authority, protect and track sensitive information, and prepare to detect attacks and restore critical systems.
• KORA says its Linux desktop can run a 26-billion-parameter model on Qualcomm’s Snapdragon X2 Elite; it expects about 70% of managed model calls to run locally in 2027.
• Token’s demonstration exposed risks when an agent acts through a user’s authenticated browser session; proposed protections still lack a shared way to convey agent identity, permissions and delegated tasks across systems.

FAA rolls out AI system to reduce delays and shorten flights

The FAA has introduced its AI powered air traffic system, SMART, nationwide to help reduce delays and find shorter routes. It analyzes 200 data streams, including weather, flight paths, air traffic and controller staffing, to predict traffic and spot possible conflicts. The FAA says SMART helped one flight from Tampa to Dallas save 42 minutes. Some lawmakers have raised concerns about testing and oversight, while transportation officials say people will remain in control of decisions.

Google selects 10 startups for its first Africa Applied AI Lab

Google has selected 10 startups from across Africa for the first Google Africa Applied AI Lab cohort. The teams work on tools for areas including farming, health care, financial services, speech recognition, satellite computing and local language access. They will get early access to Google AI models, technical mentorship and support from venture capital partners, including 4DX Ventures, Norrsken22, Novastar Ventures and Ventures Platform. The startups will work with Google experts on models such as Gemini and Gemma, then present their work at a December demo day in Accra.

A nuclear arms expert weighs Cold War lessons for AI

Keir Lieber, a professor who studies nuclear weapons and international security at Georgetown University, says Cold War nuclear arms control offers limited lessons for regulating AI. He notes that the nuclear arms race was led by the United States and Soviet Union for strategic and security reasons, while private companies drive much of AI development with a profit motive. Lieber is skeptical that countries can reach a meaningful global agreement on AI because there is no world government. He also says nationalizing the AI industry would be harder in the United States than in China, and describes AI as a new possible threat to humanity.

US and China plan AI safety talks while competing for leadership

US Treasury Secretary Scott Bessent said the United States and China should be clear about their differences as they work together on AI risks and compete for leadership. He said he will travel to China next week for President Donald Trump’s meeting with President Xi Jinping in Shenzhen on November 18 and 19, alongside the Asia-Pacific Economic Cooperation leaders summit. The two countries plan to discuss ways to test advanced AI models before release and keep human judgment in control, and have agreed to create a direct channel for reporting serious incidents. Bessent said their interests and approaches will not always match.

AI regulation emerges as a bipartisan issue after 2026 midterms

A video report examines how artificial intelligence shaped the 2026 midterm elections. It says AI regulation proved to be a bipartisan issue after the elections.

MCP and A2A Need More Than Protocols to Scale AI Agents

MCP and A2A help AI agents connect to tools, services and one another, but they do not provide everything enterprises need to scale them. Organizations must add identity and access controls, connect agents to trusted business context, redesign processes and build agentic operations capabilities. The protocols describe how agents present credentials and exchange tasks, but companies must manage identities, permissions and secrets themselves. They also need to build trust in agent results and plan how agents will work together and be monitored.

AI Security Will Shape Business Decisions in 2027

As companies move AI from trials into everyday work in 2027, security questions will become part of decisions about where and how to use it. Organizations will need to understand what information AI can access, what decisions it can affect and where people should remain involved. The article says companies should better identify and protect sensitive data, and track who accesses it and how it is used. It also urges businesses to prepare for cyberattacks by planning how to detect incidents and restore critical systems, since not every attack can be stopped.

KORA Reveals Linux Desktop With Local AI for Snapdragon PCs

KORA Systems unveiled KORA on November 10, 2026, as a Linux desktop operating system with built-in AI models and agents. The company says its engineering work with Qualcomm Technologies enables a 26-billion-parameter model to run on a Snapdragon X2 Elite chip, alongside speech recognition and embedding models. KORA expects about 70% of the model calls it manages to run locally in 2027, helping keep sensitive information on the computer and reduce cloud use. Its local context and memory stay on the user’s device, while AppArmor permissions and separate neural workspaces control access to files, tools and services. KORA is developing an SDK for third-party apps, and its Cortex system remains proprietary even though users can choose models from different providers.

Token Tests an AI Agent That Breaks Key Security Rules

Token built a browser-based AI agent that acted inside a user’s authenticated session, making its actions look like the user’s to the application. The company says the setup lacked a separate agent identity, job-specific access and traceability. Token’s Rehearsal agent can show when it is in control, record its actions and use read-only dashboards, while requiring user approval to save changes. The article points to possible protections from browsers, apps and agent identity standards, including Chrome auto-browse, Microsoft controls, WebMCP and the Web Bot Auth draft. It says these approaches still lack a shared layer to pass information about an agent’s identity, permissions and delegated task between the browser, app and agent.

Sources

• FAA turns to AI to cut delays and shorten flights nationwide
• Meet the 10 startups joining the first Google Africa Applied AI Lab cohort
• What the Cold War Nuclear Arms Race Can Teach Us About AI
• 'No Illusions' About US-China Differences Over Artificial Intelligence
• How Artificial Intelligence Shaped the 2026 Midterms
• MCP and A2A alone won’t scale AI agents
• In 2027, Every AI Conversation Will Become a Security Conversation
• KORA Unveils Linux Desktop With On-Device AI for Snapdragon PCs
• We Built An Agent That Broke All of Our Security Rules. Here's What We Learned.