When a government agency stops saying “artificial intelligence” and starts saying “automated decision-making,” or when a White House office quietly replaces “machine learning” with “trustworthy AI,” it can look like housekeeping. It rarely is. The words officials choose to describe a technology shape what rules seem appropriate, which fears get legitimized, and which industries gain or lose ground. Naming, in technology policy, is a form of governing.
The Term Shapes the Terrain
Language frames what a problem looks like before anyone proposes a solution. Consider how differently regulators approach a system described as “automated hiring software” versus one called “an AI recruiter.” The first sounds like a workflow tool subject to employment law. The second sounds like an autonomous agent that might need its own regulatory category. The underlying software could be identical.
This is not a hypothetical concern. The European Union’s AI Act, which legislators finalized in 2024, spent years debating exactly which systems would fall under its definitions. The scope of the law, and the compliance burden on companies, turned substantially on how “AI system” was defined in the text. Lobbying over that definition was intense precisely because the label determined who regulated what.
In the United States, federal agencies have cycled through terminology with notable frequency. The phrase “algorithmic accountability” gained traction in policy circles around 2019 and 2020, only to be partially displaced by “responsible AI” and later “AI safety,” each carrying slightly different implications about where responsibility sits, with developers, deployers, or regulators.
Silicon Valley’s Stake in the Vocabulary
Technology companies are not passive observers of this process. Rebranding has real commercial consequences, and the industry has long understood that softer language can soften regulatory scrutiny. When large language models became a source of public anxiety after the broad release of ChatGPT in late 2022, several companies began describing their products less as “AI” and more as “productivity tools” or “assistants.” The goal, critics argued, was to keep the products inside familiar consumer software categories rather than novel ones that might attract new rules.
The reverse also happens. Some companies lean into the “AI” label aggressively in investor materials and press releases while downplaying it in regulatory filings, because the term signals innovation to markets even as it attracts scrutiny from governments. The same product, two different names, depending on the audience.
Academics who study science and technology policy have noted for years that naming a technology shapes what gets measured, what gets funded, and what gets banned. The move from “surveillance technology” to “public safety analytics,” for example, does not change how a facial recognition system works, but it does shift the default assumption about whether its use is legitimate.
When Renaming Reflects Genuine Progress, and When It Does Not
It is worth acknowledging that terminology does sometimes change for legitimate reasons. “Machine learning” replaced older terms like “expert systems” partly because the underlying methods genuinely shifted. “Generative AI” describes a real and meaningful technical category that did not exist in mainstream commercial deployment a decade ago. Language evolves with the technology, and that is appropriate.
The test is whether the new label clarifies or obscures. A rename that helps the public understand what a system actually does serves everyone. A rename that makes a powerful, potentially risky system sound mundane or benign primarily serves the entity deploying it. The two are not always easy to distinguish from the outside, which is precisely why journalists, researchers, and engaged citizens need to ask who proposed the new terminology and what they stood to gain from it.
As governments around the world build out AI governance frameworks in the mid-2020s, the definitional battles will only intensify. The next time an official or a company announces a shift in how it describes a technology, the productive question is not what the new name sounds like. It is what the new name makes it easier to do, or harder to regulate.