US Experts: AI Regulation Benefits US, Not China

Key Takeaways

• <ul><li>Experts warn unregulated AI threatens US national security by making it easier for China to use American advances through techniques like distillation</li><li>The US and China are forming rival AI blocs, and countries like Kazakhstan are struggling to choose between them</li><li>Arize's Harbor framework showed a 43% drop in tokens and 31% cost reduction when using sandboxes instead of tool catalogs for AI agents</li><li>FireMon launched Agentic AI for network security policy, allowing agents to search data, build recommendations, and prepare coordinated changes</li><li>Oracle introduced Oracle AI Security Mesh at Oracle AI World, using OpenAI cyber models to find security weaknesses with customer approval required before fixes</li><li>IDC research commissioned by Intel shows 89% of IT leaders expect agentic AI to matter within 12 to 24 months, but only 56% say infrastructure is well funded</li><li>Former NSA operator Jay Kaplan says AI gives attackers powerful tools, enabling a single person to target many institutions using agentic tools and models like Claude</li><li>vLLM Semantic Router released d3-ultra, a 27B multimodal decision model with 26.09B parameters supporting 131,072 tokens of context under Apache-2.0 license</li><li>Kenya's second AI training cohort through UNESCO targets 9,000 public servants, with 2,380 enrolments already recorded via the Kenya School of Government</li><li>Marquette University's Raynor Library hosted a student discussion on AI on November 4, presented by the Discerning AI Research Community</li></ul>

Unregulated AI may harm US national security despite China fears

Trump has argued that regulating AI would slow US progress and help China take the lead. However, experts say unregulated AI also threatens US security by making it easier for China to use American AI advances through techniques like distillation. Chinese officials have actually called for more AI regulation, not less. The article argues that US regulations could strengthen national security by limiting China's access to cutting-edge AI.

Countries caught between US and China AI blocs

The United States and China are forming rival AI blocs called Pax Silica and the World AI Cooperation Organization. Many countries are struggling to choose between US advanced chips and Chinese open-source AI systems. Kazakhstan is one example, joining both groups while also accepting investments from both nations. Experts say the window for countries to freely choose is narrowing.

Give AI agents sandboxes instead of tools, expert says

Mikyo King of Arize advises teams to stop giving AI agents a catalog of tools and instead use sandboxes. Sandboxes offer better context management, lower costs, and more efficient token use through code mode. Testing with Arize's Harbor framework showed a 43% drop in tokens and 31% cost reduction. However, sandboxes still need careful management of networking, permissions, and debugging.

AI multiplies attacker expertise while defenders lack talent

Jay Kaplan, former NSA operator and CEO of Synack, says AI gives attackers powerful tools that multiply their effectiveness. A single person with agentic tools can now target many institutions in weeks, a task that once required skilled teams. Recent hacks showed attackers using AI tools like Claude with minimal technical skill. Defenders still face a talent gap in identifying real attack paths among thousands of machine-generated findings.

FireMon launches Agentic AI for network security policy

FireMon announced FireMon Agentic AI, a new system that helps teams investigate network access and make policy decisions. Agents can search policy data, build recommendations, and prepare coordinated changes across devices. Customers can use supplied agents, build their own, or connect external workflows through Model Context Protocol. The company says this approach saves thousands of hours while keeping human judgment over key decisions.

Oracle launches AI security mesh with OpenAI models

Oracle announced Oracle AI Security Mesh at Oracle AI World in Las Vegas. The system uses OpenAI cyber models combined with Oracle testing tools to find security weaknesses. It gives security teams evidence and test results while requiring customer approval before applying fixes. Oracle also introduced Cloud Infrastructure Identity Assurance using biometric verification and highlighted OCI Zero Trust Packet Routing.

IDC report says enterprise infrastructure must catch up with agentic AI

New IDC research commissioned by Intel shows 89 percent of IT leaders expect agentic AI to matter within 12 to 24 months. However only 56 percent say agentic AI infrastructure is well funded. The report calls for hybrid and heterogeneous infrastructure using both CPUs and GPUs rather than relying on one type of hardware. About 45 percent of respondents named modernizing data center infrastructure as a top priority.

The AI enthusiasm gap is a dangerous misnomer

This article argues that calling China's public support for AI an enthusiasm gap is misleading. It warns that framing China's AI progress as an advantage may push Americans to avoid important debate. The article also raises concerns about the Trump administration using AI tools to influence elections through disinformation. It says public pressure should not be removed from decisions that carry major political consequences.

vLLM Semantic Router releases d3-ultra multimodal decision model

vLLM Semantic Router released d3-ultra, a 27B multimodal decision model licensed under Apache-2.0. The model does not generate text but reads input once and returns probabilities for every answer. It has 26.09B parameters including a 0.46B vision encoder and supports 131,072 tokens of context. It can handle routing, classification, visual grounding and video checks in one call.

Kenya expands AI training to 9,000 more public servants

UNESCO reports that Kenya's second AI training cohort has launched. The program targets 9,000 public servants with 2,380 enrolments already recorded. The training is delivered through the Kenya School of Government. The initiative aims to build AI skills across the public sector.

Students discuss AI at Marquette University event

Raynor Library's Connect team will host a student discussion about artificial intelligence on Wednesday, Nov. 4, at 6 p.m. in the Raynor Library Beaumier Suites. The event aims to collect student views on AI use and how to discern AI. Students need their MUID for entry and do not need to register. Pizza, salad and soda will be provided. The event is presented by Raynor Library and MU's Discerning AI Research Community.

Sources

• Trump Frames Unregulated AI as a Way To Keep Ahead of China, but in Fact It Harms Us National Security
• As the U.S. and China Form AI Blocs, Countries Are Increasingly Torn Between Them
• Stop handing tools to agents, give them sandboxes
• NSA official turner founder: AI multiplies attackers’ expertise while defenders are still short on talent
• New FireMon Agentic AI Redefines Network Security Policy Management
• Oracle launches AI security mesh powered by OpenAI models
• Agentic AI Is Moving Fast. Enterprise Infrastructure Needs to Move...
• The AI “enthusiasm” gap is a dangerous misnomer
• vLLM Semantic Router Releases d3-ultra: A 27B Open Multimodal Decision Model
• Kenya expands AI training to 9,000 more public servants
• Artificial intelligence in a Jesuit University: A student conversation, Nov. 4