US Pressures Mexico for Stricter AI Hardware Rules, Exports Surge to $82.9B

Key Takeaways

• <ul><li>Mexico exported $82.9 billion in AI computer servers in the first half of 2026, up 172.1 percent year over year, with the US absorbing 93.9 percent of those exports.</li><li>The US is pressuring Mexico to adopt North American content rules for AI hardware similar to auto sourcing requirements, to prevent tariff circumvention via China.</li><li>OWASP's 2026 LLM security ranking moved Excessive Agency to third place, based on 6,639 recorded incidents, highlighting growing risks around AI agent permissions and tool access.</li><li>Canada and Germany will invest CAD 150M and EUR 100M in LawZero to build Yoshua Bengio's Scientist AI, focused on transparent, evidence-based reasoning.</li><li>Commission President Ursula von der Leyen proposed an EU Kids Act banning social media under age 13 and personal accounts under age 15 at the SOTEU 2026 debate.</li><li>Crusoe and Perplexity signed a multi-year deal using NVIDIA GB300 NVL72 clusters, which reportedly boost AI factory output by up to 50x over Hopper-based platforms.</li><li>Fleet Device Management joined Anthropic's Project Glasswing to explore how frontier AI can improve defensive workflows like endpoint assessment and patch prioritization.</li><li>Eightfold AI's Interviewer won the 2026 Top HR Product award, built on a platform trained on 1.6 billion career trajectories and 1.6 million skills.</li><li>TJGC Group appointed Kalvin Kwok to lead subsidiary Tongjiang Group's entry into AI hardware sourcing and assembly, effective September 16, 2026.</li><li>Sogang University launched a Youth Employment Academy program for AI-specialized game development, with applications due by October 5.</li></ul>

US proposes strict North American content limits for AI chips and servers

The US wants to set a hard cap on how much of an AI chip or server can be built from parts sourced outside North America. The goal is to stop Chinese firms from routing components through Mexico to avoid US tariffs. AI hardware has become Mexico's largest export to the US this year, surpassing automobiles. A new round of US-Mexico trade talks is expected soon. The proposal could reshape supply chains for manufacturers with cross-border operations.

US pushes Mexico on AI hardware sourcing rules to limit Chinese imports

The US is pressuring Mexico to accept new rules limiting foreign components in AI hardware like chips and servers. The proposal aims to prevent Chinese companies from bypassing US tariffs through Mexico. Mexico exported 82.9 billion dollars worth of AI computer servers in the first half of 2026, a 172.1 percent increase from the year before. The US absorbed 93.9 percent of those exports. The US wants AI hardware to face sourcing rules similar to autos, which require 75 percent North American content.

US presses Mexico on AI hardware rules to block tariff circumvention

The US is urging Mexico to adopt stricter rules of origin for AI hardware to stop Chinese firms from sidestepping tariffs. Trade negotiators Jamieson Greer and Marcelo Ebrard have discussed limiting non-market inputs in North American supply chains. Mexico exported 82.9 billion dollars in computer servers in the first half of 2026. Senator Bernie Moreno warned that China could use Mexico as a gateway to circumvent trade restrictions. The USMCA trade relationship remains under active renegotiation.

AI security risks shift to agent permissions and tool access

OWASP's 2026 Top 10 for LLM security moved Excessive Agency from sixth to third place. The risk is no longer just about what a model says but what tools and credentials it can access after answering. Improper Output Handling fell to tenth place. The report was the first weighted partly on 6,639 recorded incidents. Experts recommend read-only tools, scoped user identities, policy enforcement points, and human approval for high-impact actions.

OWASP ranks agent permissions as top AI security concern

Excessive Agency rose to third in OWASP's 2026 LLM security ranking while Improper Output Handling dropped to tenth. The 2026 edition was the first weighted partly on incident data, using 6,639 recorded incidents. Europe's Cyber Resilience Act requires 24-hour vulnerability reports starting September 11. The regulation covers products but may not fully apply to cloud-delivered AI agents. Security experts stress limiting agent tools, credentials, and scopes to reduce risk.

TJGC subsidiary enters AI hardware with new executive

TJGC Group announced its subsidiary Tongjiang Group will start sourcing and assembling AI hardware parts like DRAM, SSDs, CPUs and integrated circuits. Kalvin Kwok, a veteran with over 30 years in electronics manufacturing, was appointed executive director effective September 16, 2026. TJGC will transfer 49% of Tongjiang to Kwok after his first binding purchase order, but TJGC keeps full control of operations and finances.

Kalvin Kwok to lead Tongjiang's AI hardware business

TJGC Group Limited appointed Kalvin Kwok as executive director of its Hong Kong subsidiary Tongjiang Group Limited, effective September 16, 2026. Kwok will lead Tongjiang's new business in sourcing and assembling high-performance PC and data-center components to meet growing global AI infrastructure demand. He brings over 30 years of experience in electronics manufacturing. Kwok will receive 49% of the subsidiary's shares after securing its first purchase order, while TJGC retains full operational control.

Sogang University launches AI game development training

Sogang University and the Seoul Anchor Center are running a Youth Employment Academy program for AI-specialized game development talent. The program offers two courses: an Education Course with 24 participants and 286 hours, and a Project Course with 28 participants and 344 hours. Both courses cover generative AI game development with hands-on projects and mentoring from industry developers. The application deadline is October 5 at 11:59 PM.

Eightfold AI Interviewer wins 2026 Top HR Product award

Eightfold AI's Interviewer was named a 2026 Top HR Product of the Year by HR Executive. The agentic interviewing platform is built on Eightfold's Talent Intelligence Platform, trained on 1.6 billion career trajectories and 1.6 million skills. It supports structured, skills-based conversations for screening, coding and language use cases. The award was given after evaluation by a panel of judges and a live product demonstration.

MEPs debate economy, AI, climate and social media rules

At the SOTEU 2026 debate in Strasbourg on September 16, 2026, MEPs and Commission President Ursula von der Leyen discussed the EU's key challenges. Von der Leyen said boosting computing capacity and investing in AI start-ups are priorities, along with climate resilience and a European security council. She also proposed an EU Kids Act banning social media under age 13 and personal accounts under age 15. The Commission plans to tighten sanctions on Russia and support Ukraine.

Crusoe and Perplexity partner on AI training and inference

Crusoe announced a multi-year deal with Perplexity on September 15, 2026. Perplexity will run its full model lifecycle on Crusoe Cloud using dedicated NVIDIA GB300 NVL72 clusters. The deal also has Crusoe adopting Perplexity products for its 1,800 employees. NVIDIA's GB300 NVL72 is said to boost AI factory output performance by up to 50x compared with Hopper-based platforms.

Fleet Device Management joins Anthropic's Project Glasswing

Fleet Device Management joined Anthropic's Project Glasswing to improve AI security for device management systems. The project explores how frontier AI can strengthen defensive workflows like endpoint assessment and remediation prioritization. A survey by Fleet found that 79% of organizations take over a day to deploy key security patches. Anthropic launched Project Glasswing in April 2026 to address concerns about AI and vulnerability discovery.

Canada and Germany fund Bengio's safe AI initiative

Canada and Germany will invest CAD 150M and EUR 100M in LawZero to build Yoshua Bengio's Scientist AI. The AI is designed to reason transparently and produce evidence-based outputs without pursuing its own goals. Canada's share will create 360 full-time jobs and build sovereign computing infrastructure with Canadian firms Hypertec and 5C. Germany's funding will support R&D at LawZero's new German office and incorporate European norms.

Will an AI slowdown tank the economy?

The article discusses the tension between AI companies that want to slow down model building and investors who want faster growth. Frontier developers like Anthropic and OpenAI face financial pressure to pace their model-building. Slowing down could help Anthropic's financial disclosures before a public offering. However, infrastructure suppliers like Nvidia, Oracle, and Broadcom cannot easily handle a slowdown without losing revenue.

Sources

• Washington seeks cap on foreign parts in AI chips and servers
• GT Voice: Will US sourcing rules shock AI hardware supply chain?
• US Presses Mexico to Tighten AI Hardware Rules to Block China From Circumventing Tariffs: Report
• The Hardest AI Security Problems Now Live Outside the Model
• The hardest AI security problems now live in what an agent is permitted to do
• New TJGC (TJGC) exec set to gain 49% of AI hardware arm
• TJGC Group Limited Appoints Kalvin Kwok to Lead Tongjiang Group’s AI and Data-Center Hardware Business
• Sogang University Operates AI-Specialized Game Development Talent Training Program
• Eightfold AI Interviewer Named a 2026 Top HR Product of the Year by HR Executive
• SOTEU 2026: MEPs debate economy, security, climate, AI and social media
• Crusoe Signs Multi-Year Deal to Power Perplexity Training and Inference
• Fleet Device Management joins Anthropic’s Project Glasswing to enhance AI security
• Canada and Germany fund Bengio's safe AI push
• Will an AI Slowdown Tank the Economy?