The Accountability Gap: When AI Policy Becomes Weapon Policy

The Policy Framework That Eats Itself

The Trump administration's AI Action Plan, signed July 2025, had three stated goals: accelerate AI innovation, build American AI infrastructure, and lead international AI diplomacy. What it actually did was centralize decision-making about AI deployment in the federal government while stripping away the oversight mechanisms designed to catch errors.

An executive order created a DOJ task force to litigate against state AI regulations and conditioned federal funding on states pausing enforcement of their own AI laws. A proposed 10-year moratorium on new state AI regulations followed. Sensitive government data — including Americans' private information — was fed into unvetted AI software. An unapproved AI chatbot was deployed onto Department of Homeland Security systems.

The tool at the center of this was SweetREX Deregulation AI, created by Christopher Sweet, a former University of Chicago student and DOGE affiliate. It uses Google AI models. It wrote "100% of deregulations" at the Consumer Financial Protection Bureau. The administration's own internal PowerPoint divided all federal regulations into three buckets: half not required by law, 38% statutorily mandated, 12% "Not Required but Agency Needs." The goal: eliminate 100,000 rules by the first anniversary of Trump's return to office, labeled internally as "Relaunch America."

The Personnel Problem

Among the young engineers Elon Musk hand-picked for DOGE was Edward Coristine, 19, known online as "Big Balls." He dropped out of Northeastern University, briefly worked at Musk's Neuralink, and was previously fired from an internship at a cybersecurity firm for leaking company secrets. His company DiamondCDN's services were allegedly used by a cybercriminal gang. At 16 he registered Tesla.Sexy LLC, which manages dozens of web domains including Russian-registered sites — one running an AI bot operating in Russia.

A former FBI agent said he would not have recommended Coristine for government work. He was given senior adviser roles at the State Department's Bureau of Diplomatic Technology and at CISA — the nation's top cybersecurity agency. He and another DOGE associate promoted AI deployment across the federal bureaucracy. He moved from DOGE to GSA to the Social Security Administration before eventually resigning. At no point was his access to Americans' sensitive government data publicly questioned by the administration that granted it.

The Quantum Money Pipeline

In June 2026, two executive orders were signed in the Oval Office alongside Alphabet's president, IBM's CEO, and a Nobel Prize-winning physicist. The first promotes commercialization of quantum computing. The second mandates nationwide migration to post-quantum cryptography.

The justification: "harvest now, decrypt later" — adversaries are already intercepting and storing today's encrypted communications, waiting to decrypt them once quantum computers capable of breaking current encryption become available. The 2030 and 2031 deadlines are not preparation for a future problem. They are a response to an attack happening now.

This is the same pattern at a different scale: centralized implementation, billions in public funding, conflicts of interest, no transparency about the conceptual errors already built into the systems this infrastructure will run on — deployed by the same administration that used an AI tool that hallucinated contract values to rewrite the rules that govern people's lives.

The Chip Chokepoint

Every system documented in this article — the AI deregulation tools, the quantum infrastructure, the federal AI governance — physically depends on chips. Nvidia designs them. TSMC in Taiwan fabricates them. Nvidia doesn't manufacture anything. It outsources to the island China has stated it intends to reunify with by force if necessary.

Nvidia's own chief scientist, Bill Dally, publicly stated that former Nvidia engineers in China are now working for Huawei. The US ban on the H20 AI accelerator gave Chinese firms room to grow and helped them seize high-end AI talent. In 2019, China's AI researchers were less than a third of the global count. Now it's 50%.

While the US government spends hundreds of millions trying to keep chips out of China, Nvidia is simultaneously recruiting autonomous driving talent in Beijing, Shanghai, and Shenzhen — a team reporting directly to Jensen Huang. The boundary isn't holding. The chips are flowing. The talent is flowing. The technology is flowing.

The Trillionaire Entanglement

In June 2026, Elon Musk became the world's first dollar trillionaire when SpaceX went public at a $1.8 trillion valuation. Jensen Huang is at $173–200 billion, with prediction markets giving him 50% odds of being the next trillionaire.

Musk's xAI leases capacity on the Colossus cluster — over 220,000 Nvidia GPUs — paying $1.25 billion per month through May 2029. The first trillionaire's AI empire runs on the chips of the man most likely to be the second trillionaire. Musk is simultaneously Huang's premier customer and his primary competitor in frontier AI training.

Meanwhile, Charles Liang — Supermicro's CEO, whose cofounder was arrested for a $2.5 billion China smuggling scheme — announced in June 2026 that Supermicro will build a new gigawatt-scale AI data center for SpaceX and xAI, with a completion timeline faster than any previous build. The same company linked to massive chip diversion is now building critical national security infrastructure for the world's first trillionaire.

The Accountability Vacuum

A January 2026 study from the Wharton School named the mechanism: cognitive surrender — adopting AI outputs with minimal scrutiny, overriding both intuition and deliberation. MIT brain imaging research found measurable neural under-engagement during and after LLM-assisted work. The networks responsible for focus, memory, and attention showed reduced activity.

But cognitive surrender isn't just an individual phenomenon. It's a systemic one. When AI tools are deployed to rewrite federal regulations, when quantum infrastructure is funded through channels with direct administration ties, when the world's most valuable company operates on both sides of a geopolitical divide simultaneously — the accountability mechanisms that would normally catch these conflicts are themselves being replaced by the systems that create them.

The US government has made Nvidia the single point of failure for its entire AI infrastructure — military, intelligence, and civilian. That same company depends on a Taiwanese manufacturer in the most geopolitically contested region on Earth. Its CEO is a presidential advisor lobbying to sell advanced chips to China while his chief scientist warns that export controls are accelerating Chinese AI development. And a company whose leadership has been linked to a $2.5 billion China smuggling scheme is now building the infrastructure for systems with direct defense applications.

The Bifurcation Point

Iran shut the Strait of Hormuz this year without a navy. A handful of missile and drone strikes convinced insurance markets the risk was too high. Commercial shipping stopped on its own. No formal blockade needed. The chokepoint closed itself.

China watched and learned. Military analysts are calling it Hormuz 2.0. China has already conducted boarding inspections in the Taiwan Strait under the pretext of maritime inspections. The same chokepoint strategy — applied to the strait that over 90% of the world's most advanced semiconductor chips pass through.

Oil through one strait, chips through another. Energy and computation. Both passing through a single boundary where everything converges and everything could stop. Two chokepoints for everything the modern world runs on.

The trajectory that held — open straits, flowing supply chains, distributed access to processing power — is becoming unstable. New attractors are appearing. Closed straits. Concentrated access. Systems that run on chips manufactured in facilities that a single military action could sever from the global supply chain.

The Question That Remains

What happens when "thinking" is done by systems only a very small group of people have access to? The AI tools were installed to think for government — scan the rules, make the decisions, replace the work hours. Now quantum systems are being installed to protect the government from threats that may have been introduced by the same people installing the protection.

Each layer replaces more human thinking with systems that operate further from human oversight. Each step is justified as protection. But the protection is being built by the same structures that created the vulnerability. And the accountability for what happens when those systems fail — when they hallucinate contract values, when they misread statutes, when they rewrite the rules that govern billions of lives — is being offloaded to mechanisms that process without awareness of what they're processing.

The consequences are not artificial. They are the reality for billions of people who live under systems they did not choose, governed by decisions they cannot see, running on infrastructure that could stop with a single event at a single strait. Gaza and Iran are the current examples. They will not be the last.