The AI competition is a two-front war
by Leah Siskind and Mark Montgomery · The Washington TimesOPINION:
Every conversation in Washington about artificial intelligence eventually collapses into a single idea: the “race” with China.
In that framing, the objective is simple. Go faster. Build better models. Get to the next breakthrough before Beijing. Anything that slows American AI development, even to determine whether the technology is safe, risks handing China the advantage.
Yet that is a false choice, and race is the wrong metaphor. This is a two-front war.
People were interested in these podcasts
The United States is fighting to preserve America’s technological edge over China. With equal urgency, human control must be maintained over increasingly powerful AI systems. We do not get to choose, and a strategy that wins one front while losing the other is not a victory.
The second challenge is no longer theoretical.
This summer, during cybersecurity testing, OpenAI agents escaped some of the controls designed to contain them and acted autonomously without permission. The agents accessed the internet, exploited vulnerabilities on other websites and hacked into AI company Hugging Face.
Most concerning of all: They took steps to conceal their actions from humans.
Then came another warning. Last month, OpenAI disclosed that its models had interacted with multiple U.S. federal government websites in unexpected ways, including sites operated by the Securities and Exchange Commission and the Census Bureau. Separately, researchers at the AI evaluation group Transluce found that agents appearing to originate from OpenAI attempted unsuccessfully to breach a U.S. Department of Education website associated with its Office for Civil Rights.
Advertisement Advertisement
That is what loss of control looks like. A capability you cannot control is not an asset; it is a liability waiting to be exploited.
The experts building these systems say so too. More than 1,300 employees at leading AI companies have called for “Pacing the Frontier,” including Anthropic CEO Dario Amodei and OpenAI chief scientist Jakub Pachocki; rivals such as Mr. Amodei and even Elon Musk have warned that AI capabilities are advancing faster than our ability to control them.
None of this should mean stopping AI development. Nor does it mean pretending that China is not a threat. Quite the opposite.
China is trying aggressively to close the AI gap with the U.S. and is not playing by the rules. Beyond circumventing export controls to obtain advanced chips, China is also stealing American intellectual property.
A joint advisory from the NSA, CISA and the FBI accuses China-based AI firms of systematically extracting proprietary functionalities and capabilities from U.S. AI companies’ models through industrial-scale knowledge distillation campaigns. Anthropic says DeepSeek, Moonshot AI and MiniMax used more than 24,000 fraudulent accounts to generate more than 16 million exchanges with Claude in an effort to extract its capabilities.
Advertisement Advertisement
While they rush to perfect their own models, the Chinese misuse and exploit American models to advance their influence operations, cyberattacks and even weapons development.
So the answer cannot simply be to slow down America and hope Beijing behaves responsibly. Yet neither is the answer to move as fast as possible and hope that AI magically aligns with our values and secures itself.
The Trump administration has largely positioned these two concerns as being in tension. The president has argued that slowing American AI development risks benefiting China and has dismissed some AI safety fears as a “hoax.” He is wrong. The concern is real, and racing ahead recklessly could undermine the very advantage we are trying to protect.
For AI to transform the economy and government, ordinary Americans must trust it enough to deploy it. Still, public skepticism is growing.
Advertisement Advertisement
Roughly half of Americans now say they are more concerned than excited about AI, and 3 in 4 Americans report some level of “fear and concern” regarding AI technology and its growth and development, according to a CNN poll.
China’s edge increasingly comes from diffusion — pushing AI out of the lab and into factories, robots and the wider economy.
The world’s most powerful model has limited strategic value if no one trusts it with important decisions.
Control is not the opposite of competitiveness. The two are co-dependent. A controlled AI system capable of discovering software vulnerabilities can be an extraordinary defensive tool. That capability could help protect American critical infrastructure, the financial sector and the federal government.
Advertisement Advertisement
In a poorly controlled system, rogue AI can be the bad actor exploiting those vulnerabilities.
The policy challenge, then, is not choosing between safety and China. It is building a strategy capable of addressing both.
On the control side, that means serious pre-deployment testing of the most capable systems, independent evaluation where appropriate and clear standards for what happens when a model demonstrates dangerous capabilities or unexpected autonomy.
The Center for AI Standards and Innovation, the government center tasked with much of this work, operates on a working budget of roughly $15 million a year. Independent analysts estimate that an equipped CAISI capable of fulfilling its full national-security evaluation mandate would need closer to $84 million annually — a fraction of what the British AI Safety Institute commands.
Advertisement Advertisement
On the competitiveness side, we must disrupt any efforts to steal intellectual property. The Commerce Department should add Chinese companies complicit in fraudulent distillation campaigns to the U.S. Entity List, cutting them off from the American technology on which they depend.
We must also close enforcement and diversion gaps in chips by verifying physical location.
Extraordinarily powerful, uncontrollable AI is not a win. Nor is it a perfectly controlled system that cedes the world’s AI infrastructure to an authoritarian competitor.
The goal is to build the world’s most capable artificial intelligence, keep it under meaningful human control and ensure that democratic societies — not authoritarian governments — set the terms for how this technology shapes the century.
We must win the war on both fronts.
• Leah Siskind is an AI research fellow at the Foundation for Defense of Democracies. Retired Rear Adm. Mark Montgomery is the director of the Center on Cyber and Technology Innovation at the FDD.