Eye on Policy

Six hundred years ago, Henry VIII broke from Pope Clement VII. Today, when it comes to artificial intelligence at least, latter-day King Charles III and Pope Leo XIV share many ideas, including the idea that AI systems must be governed in such a way as to protect humanity. Last month, Charles hosted a sort of summit of AI leaders from industry. But he also included the Pope’s AI advisor, Fr. Paolo Benanti. When kings and popes have AI advisors, you know something big is going on.

This is all relevant to the United States and to the next generation of digital government, which will be all about AI. Notwithstanding warnings that AI will end humanity, every federal tech vendor touts its AI prowess.

It is almost impossible to keep up with the stream of news in the world of AI. Daily advances occur in the products themselves, in the corporate development and finance realm, and in political debates over whether and how AI should be regulated. That last question has huge implications for an emerging industry with multiple trillions of dollars in market capitalization, and when millions of individuals have freely brought AI into their personal lives.

The operational, bureaucratic side of government has the challenge of ensuring systems and services are up to date, while Congress, the White House and a million interest groups debate how to deal with the technology.

The takeaway for agencies and contractors: Do not deploy AI in a way that puts agencies into the debate itself. Do not avoid AI, just do not get too ambitious with it.

You can find countless limited but useful cases that do not outrun privacy, confidentiality, or security concerns. As a motorcycle rider, when riding in the dark I try to follow a simple rule: Never outrun my headlight. One deer popping out of the darkness could be a disaster. So should it be with agency deployment of AI.

A company called Flock provides a useful analogy. It manufactures surveillance cameras and systems such as flying drone-mounted cameras, all for law enforcement and public safety applications. The highly capable systems have sparked heated privacy controversy in many cities and towns that have deployed them.

The Hill reported on one group that snatched a Flock camera, dissected it, and sent the results to journalists. That group referred to the car license plate reading technology as “spying on us.” Other published reports show that grassroots citizen opposition to the installations are bipartisan. Republican Texas Governor Greg Abbott went as far as to order a freeze on installations of Flock and Flock-like technology on the state’s roads, although it leaves nine hundred cameras in place. In August, Democratic Connecticut Governor Ned Lamont urged the state police to review its use of the technology. Now the District of Columbia Metropolitan police are looking into misuse of Flock data amid concern by the city council and some citizens.

To be sure, the highly capable technology has solved or prevented crime in numerous instances. But when technology sparks serious and sustained opposition, loss of faith in it by elected leaders often follows.

Artificial intelligence presents an even more complicated problem. An agency can disconnect cameras, but AI models have shown repeatedly they can act unintended by their operators and acquire autonomy.

To try and summarize what is going on:

• Justifiably or not, AI has become a mainstream “dinner table” topic, after an under-30 Anthropic researcher quit the company, warning of AI killing everybody. The media took up the warning as gospel, so now that possibility has become an assumption.
• Highly publicized AI “breakouts” have proven the technology’s destructive potential, especially in the cybersecurity domain. That is, there is plenty of risk now, never mind what could happen in 10 years.
• U.S. companies’ major competition, the Chinese, operate under governmental systems with differing views on AI. That hinders the possibility of a universal agreement on a regulatory framework.
• The Trump administration wants a light regulatory touch. President Trump, at the urging of his own AI advisor and over those of others in the inner circle, nixed the idea of having new models of Claude and the like subject to governmental inspection. Trump’s new AI task force seems aimed at promoting AI growth, rather than establishing a regulatory framework.
• Some in Congress advocate for more federal oversight. For example, Sen. Mark Warner (D-Virginia) cited the Open AI Hugging Face incident in calling for a sandbox testing environment for new models before they are released.
• EU nations—including the U.K.—have clearinghouses of their own, but they have no deployment authority in the U.S. or China. Warner, in a PBS interview, held up the British set-up as an example of how the U.S. should proceed.

Both the safety and economic growth arguments have merit. Meantime, uncertainty results from the non-emergence of coherent and accepted policy.

Technical approaches to containment and, if necessary, stoppage of AI operations exist. It takes people in the government dedicated to AI deployment to understand and invoke these safeguards. AI has a ying-yang quality. It can make great code, for example, then find the security vulnerabilities in that code.

Despite all this, AI remains compelling to agencies.

Back in June, for example, the Federal Aviation Administration hired Air Space Intelligence (ASI) in a 12-year, $875 million contract to support two programs. The FMDS and SMART programs have to do with seeing, predicting, and dealing with the flow of aircraft under FAA control. The company’s AI software, it says, is already used by the Air Force and many airlines. The FAA’s idea is to have early insight into the day’s flow. In the company’s words, “the FAA is expected to gain enhanced ability to anticipate and manage disruptions before they move through the airspace system.”

In using AI-enabled software for a specific mission-related operating purpose—and in using software proven in other situations—the FAA will presumably improve its ability to something it needs to do; namely, anticipate traffic so it can better anticipate safety issues, workloads and staffing needs.

Modernization will involve AI in many cases. That genie is out of the bottle. But contractors and agencies will need to work together to make sure it stays fully auditable, keeps humans in the loop, and remains contained on specific tasks so avoid outrunning oversight and public acceptance.