World leaders are beginning to view American AI less as a commercial product and more as a piece of critical infrastructure. During the recent G7 summit, officials from nations like France and India expressed concern about their dependency on U.S.-based models, specifically fearing that American companies could abruptly restrict access to the technology.
This anxiety coincides with a notable decline in domestic optimism, as a recent Pew Research study found that only 16 percent of Americans believe AI will have a positive impact on society. Meanwhile, the White House is pushing developers to eliminate "jailbreaks"—methods used to bypass safety protocols—but security experts warn that creating a perfectly tamper-proof model remains a persistent technical challenge.
The limits of software control
Think of a security guardrail in an AI model like a lock on a door; you can make the lock more complex, but a talented enough intruder will eventually find a way to pick it. Models like those from Anthropic rely on layers of software instructions meant to prevent prohibited output, but these instructions exist within the same neural network that allows the model to function. Because these systems are probabilistic rather than purely rule-based, they will always have unpredictable edges that can be manipulated.
For policymakers, the goal is to treat AI as a regulated utility, but the underlying technology may not be stable enough to support that kind of governance yet. If the experts are right and these systems cannot be made immune to manipulation, then regulators and the public are essentially betting on a technology that is inherently insecure. The bottom line is that until we reconcile the difference between our desire for total control and the reality of how these models function, the distrust seen in public polling is likely to stay.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy