The contours of a possible deal to keep TikTok running in the U.S. are emerging: a group of American investors including the VC firm Andreessen Horowitz, private equity firm Silver Lake, and cloud service company Oracle will reportedly be taking over U.S. operations of the short form video app, which was set to be banned here unless its Chinese parent company Byte Dance agreed to divest.

One sticking point in the negotiations has been secrecy around TikTok’s proprietary algorithm — the secret sauce behind the “For You” feed that keeps users glued to the app. A licensed copy will reportedly be run by the U.S. joint venture. The algorithm is actually considered a restricted technology under Chinese law, meaning it can’t be exported to the U.S. without special permissions. This kind of export control on an algorithm is unique, but it’s not without precedent.

Export controls have come up a lot in recent years: The U.S., for instance, has restricted the sale of advanced semiconductors to China, which, in turn, limited exports of critical minerals to the U.S. These are goods that can be used to build strategic technologies.

An algorithm, though, is a bit more abstract than elements dug out of the earth, said Steven Weber, a professor of the graduate school at UC Berkeley’s School of Information. 

“People say, ‘Oh, the algorithm,’ as if the algorithm is like a thing, or a widget, or like a something you could put in a box or drop on your foot,” he said. “An algorithm is none of those things.”

Essentially, an algorithm is just math, he said — a set of rules or instructions. But when we talk about AI, those rules are expressed in code and can self-adjust.

“It's a machine-learning system,” Weber said. “And that's made up of rules that are constantly evolving based on how people interact with it.”

And Weber said it’s a lot harder to control than a shipping container full of graphite. “Trying to control the export of math is really hard. It's like trying to nail Jell-O to the wall.”

But that hasn’t stopped governments from taking a swing at it.

Susan Landau, a professor of cybersecurity and policy at Tufts University, pointed to U.S. attempts to control encryption systems.

“Encryption systems are what we call dual-use technologies,” she said. “They can be used both by the military as well as by civilians.”

Think of the 2014 World War II movie, “The Imitation Game,” which depicts the mathematician Alan Turing’s efforts to develop computer systems to help the British government decode Nazi communications.

During the Cold War, the U.S. classified strong encryption systems literally as munitions — they couldn’t be exported without special permission. 

Landau said that became a problem in the 1990s with the rise of personal computing and the internet.

“Virtually every time you connect to a website, you're doing a secure connection,” she said. “And the kinds of things that people wanted to be able to do, in particular to do commerce, became much more complicated.”

With export controls in place, companies like Microsoft and Netscape (remember them?) couldn’t sell browsers with strong encryption outside the U.S.

“U.S. companies are really loath to develop one system with weak encryption that can be exported, and one system with strong encryption domestically,” Landau said. “It's very hard to say to customers, ‘Oh, it's fine that you have the weak encryption.’”

By the late 1990s, the government began easing restrictions on encryption systems.

But we’ll likely see more battles over export controls like this in the AI age, said Samm Sacks, a senior fellow at Yale Law School.

“We're seeing more use of it as software applications increasingly become subject to geopolitical fights,” she said.

The Biden administration proposed a framework to restrict details on the inner workings of the most advanced AI models from going overseas. The Trump administration rescinded that rule and has promoted the idea of AI as an important export, but it’s also stepped up restrictions on AI hardware for adversaries.

“At one point, it was much more clear-cut. How do you draw the line in terms of access to technology by so-called adversaries?” she said.

Now, it’s a bit like … nailing Jell-O to the wall.