September 22, 2026
The United States and China are moving quickly to develop increasingly powerful artificial intelligence systems, and Lt. Gen. Jack Shanahan (Ret.), former director of the Pentagon’s Joint Artificial Intelligence Center, says that competition is also creating new risks that both countries need to understand.
During an interview on Fed Gov Today, Shanahan discusses his message ahead of an emergency hearing of the House Select Committee on China focused on the AI competition between the two countries.
Shanahan says the United States and China remain deeply suspicious of one another, but mutual vulnerabilities surrounding advanced AI could provide a reason for continued dialogue.
He does not expect those conversations to produce a sweeping agreement. Instead, Shanahan describes a process that could take years and produce incremental areas of agreement.
One potential starting point is identifying the risks associated with increasingly capable AI models.
Shanahan says the most advanced large language models are developing at a rapid pace, with capabilities that can surprise even the companies building them. As AI agents become more capable of independently pursuing objectives, testing the boundaries placed around those systems becomes increasingly important.
Part of the challenge, he says, is defining not only what an AI system should do, but also what it cannot do.
That becomes particularly important when government takes commercially developed AI models and places them into operational or military environments.
Shanahan says agencies need additional evaluation and governance measures to understand how those systems could behave in specific operational contexts. He expects that demand to drive greater investment in modeling, simulation, red teaming and other forms of testing.
The U.S.-China relationship adds another layer to that challenge.
Shanahan says the two countries could eventually explore common technical standards for evaluating advanced models. An independent organization could potentially play a role in that process, allowing each country to assess risks without relying solely on the other side’s companies or institutions.
He also draws lessons from Cold War arms-control negotiations, while cautioning against describing AI competition as a traditional “arms race.”
Instead, Shanahan calls it a “continuous competition” without a clear finish line.
He points to the 1986 Reykjavik summit between President Ronald Reagan and Soviet leader Mikhail Gorbachev as an example of negotiations that initially fail to produce a major agreement but still create opportunities for future progress.
AI, however, presents a different challenge because much of the technology is being developed by private companies rather than governments.
That means technology companies and technical experts need to play a role in future discussions, Shanahan says.
Oversight does not necessarily have to come at the expense of innovation, he adds. Shanahan points to aviation as an example of an industry where regulation, private-sector action and independent oversight work together to improve safety.
For AI, he sees a similar need for testing and verification.
Rather than beginning with trust, Shanahan suggests starting with verification and allowing trust to develop over time.
As AI capabilities continue to advance, he says the goal is to establish technically defined ways to understand and evaluate risks while recognizing that the United States and China will continue competing to develop the technology.