GRAS works exactly as intended. That’s the problem.
The framework was originally created to evaluate isolated, time-tested substances – salt, vinegar, basil – not modern industrial formulations made from dozens of interacting components.
From an engineering standpoint, GRAS tests ingredients one at a time. And ultra-processed food behaves as a combined system. Those are not the same thing.
GRAS Asks a Narrow Question: Is This Ingredient Safe on Its Own?
Modern food creates effects only when ingredients interact – emulsifiers affecting the gut lining, flavor enhancers increasing how much we eat, sweeteners altering metabolic responses when paired with refined carbohydrates.
You don’t see those effects by looking at ingredients individually.
They only appear when components are combined, optimized, and consumed repeatedly across a population.
The regulatory gap widened in 1997, when GRAS shifted toward voluntary notification and self-affirmation. That change didn’t require bad actors. It simply assumed that evaluating individual parts would remain sufficient as food systems became more complex.
That assumption no longer holds because in complex systems, safety isn’t something you get by adding up safe parts. It emerges from how those parts interact over time.
An ingredient can be safe in isolation and still contribute to harm when stacked, optimized, and widely consumed. Other engineering fields account for this by testing how systems behave under real-world conditions. Food regulation, historically, didn’t need to.
Until now.
What’s Changed Isn’t GRAS Itself
It’s the ability to see the whole system clearly.
Once processing level, formulation patterns, and population exposure can be measured across hundreds of thousands of products, the limits of ingredient-by-ingredient evaluation become obvious – not morally, but mechanically.
GRAS wasn’t designed to fail, it was designed for a simpler world. The moment food became industrially optimized, the model aged out. And it’s time to address that.