The China Loophole Nobody Closed
The day's biggest thread is a widening gap between US export policy and US business practice. The Financial Times reports that OpenAI and Google have been supplying their advanced AI models to Singapore-based subsidiaries of three Chinese heavyweights — Alibaba, Baidu, and Tencent. A separate FT-sourced account goes further, saying the two American firms sold models to Chinese companies the Pentagon has placed on its blacklist. However the details settle, the pattern is the same: overseas arms can serve as a side door around restrictions meant to keep frontier AI out of rivals' hands. Expect this to sharpen the argument that export controls written for chips don't map cleanly onto software and cloud-delivered models.
Trust, Backdoors, and Mutual Suspicion
The distrust runs both ways. China has escalated warnings about an alleged "backdoor" in Anthropic's Claude Code, publicly flagging security vulnerabilities in the AI coding tool. Anthropic is firmly pushing back, rejecting the claim outright. The episode is a preview of how AI tools become geopolitical flashpoints — where a security allegation doubles as leverage.
American anxieties mirror it. A Duke University professor, cited via 24/7 Wall St., warns that China could remotely disable self-driving cars across the United States, colliding national-security fears with a fast-moving autonomous-vehicle industry. Yet dependence cuts both ways: the South China Morning Post notes China's AI and research ambitions still run on imported precision equipment rather than home-built gear — a hidden vulnerability in Beijing's own supply chain.
The Pentagon's AI Reality Check
Washington's military-AI drive is long on ambition and short on cash. An analysis titled "The Pentagon's AI Strategy Has a Funding Problem" argues that paying for sweeping AI plans may prove far harder than announcing them. That tension frames the day's contracts and bets.
Accenture landed an $821 million task order to integrate the Defense Department's "War Data Platform" — a rebrand of the data effort formerly known as Advana, per DefenseScoop. Elsewhere, the Pentagon is putting money behind directed-energy weapons, awarding deals for laser systems built to fry swarms of attacking drones. And AI is creeping into the unglamorous core of readiness: defense agencies are deploying it to forecast staffing needs — who's needed, where, and when.
All of it feeds a governance debate now intensifying: as militaries fold AI into weapons, logistics, and decision-making, how should these systems be governed — and should fear be the emotion steering the rules? It's the question sitting underneath every contract above.
On the Cyber Front
One quieter but telling move: cybersecurity firm Tenable has joined OpenAI's Daybreak Cyber Partner Program, formalizing a tie between an established security vendor and a frontier lab. As AI becomes both a weapon and a target, these partnerships are where the defensive playbook is being written.
The throughline: frontier AI is now a national-security asset whose controls leak, whose trust is contested on both sides of the Pacific, and whose military promise is bumping into a very ordinary obstacle — the budget.