The Permitted List

There is a sentence on the Food Standards Agency's consumer page about additives, and it is doing more work than any seven words on a government website should have to.

"A food additive is only approved if; it has been tested and proved to be safe for its intended use."¹

Read it once and it closes the question. Tested. Proved. Safe. The eye lands on the word "safe," registers a verdict, and moves on. That is what the sentence is for, and on its own terms it is true.

But look at the four words quietly holding the whole thing up: for its intended use. Those words are the scope. They tell you that the safety here is the safety of the thing doing its job — emulsifying, stabilising, keeping a loaf soft for a week. They do not say "safe for your children to eat habitually, for years." They do not name an endpoint. They do not name a duration. And they certainly do not tell you which of the dozen additives in a typical weekly shop sits inside a scientific question that is open right now, being actively worked on by the regulator itself.

This is a report about that gap — not a gap in the chemistry, but a gap between what the word certifies and what the shopper reads into it. It is the kind of gap that survives no matter how the science finally lands, because it lives in the word on the shelf, not in the body.

What "approved" actually certifies

Start with the true, narrow thing, because it is genuinely reassuring and the report depends on you believing it.

When an additive is "permitted" in Great Britain, it has passed a prescribed evaluation. Under the framework inherited from Regulation (EC) No 1333/2008, an additive earns its place by clearing three tests: it must be demonstrably safe at expected intake, it must do a justifiable technological job, and its use must not mislead the consumer.¹ The toxicology behind that first leg is real work — not a rubber stamp. The workhorse study is the repeated-dose 90-day oral toxicity assay, OECD Test Guideline 408, and its endpoint list is published and fixed: body weight, food and water consumption, haematology, clinical biochemistry, urinalysis, organ weights, gross necropsy, and full histopathology of preserved tissues — including the stomach and the small and large intestines.²

There is also a number behind the word, for most additives. The Acceptable Daily Intake, or ADI, is a toxicity ceiling — derived from the highest dose at which no adverse effect was observed, divided by a safety factor. For polysorbate-80 (E433) the group ADI is 25 mg per kilogram of body weight per day.³ For carboxymethylcellulose (E466), the ADI is literally "not specified" — because the polymer is not absorbed; it passes through and leaves intact, so the regulator concluded there was "no need for a numerical ADI" and "no safety concern at the reported uses and use levels."⁴

This is the part the panic gets wrong. The folk belief that "E-number equals poison," that if you cannot pronounce it, it must hurt you — that belief is uncoupled from the evidence. The approval system genuinely screens for acute and sub-chronic toxicity, and most of the thousands of approved additives carry no emerging chronic-health signal at all. If you finish this report believing your bread is dangerous, the report has failed and so have you. That is not where this goes.

It is worth saying plainly what "approved" is: a real, meaningful safety signal, earned by a real evaluation. Hold that. Now look at the one thing it was never built to carry.

The assay, and the unit it does not measure

Run a finger down the OECD 408 endpoint list and you will find the gut. The protocol requires histopathology of "the stomach, small and large intestines (including Peyer's patches)."² The tissue is looked at. A pathologist scores it for damage — lesions, inflammation visible under a microscope, gross structural injury.

But search the full guideline for the word "microbiome." Zero results. Search for "microbiota." Zero. Search for "mucus." Zero.²

This is the precise hinge of the whole report, so it is worth slowing down. The gut tissue is examined for damage. It is not examined for the unit that the emulsifier research actually moves: the composition of the bacterial community in the lumen, the thickness of the protective mucus layer that sits on top of the tissue, the distance maintained between the bacteria and the cells beneath. A 90-day study can return a clean "no adverse effect on the gastrointestinal tract" by histopathology while those things shift, because the protocol does not stain for them. The microbiome lives on top of the tissue. The test reads the tissue.

Here is the part that matters most, and the part this report most needs you to read as a boundary and not as a fright: when no one has asked a question, the answer that comes back is not "safe." It is silence. A study that never measured your gut microbiome cannot report a finding about your gut microbiome — and "no finding" is a very different object from "a finding of safety." They feel identical in the mind. They are not the same thing. The honest reading of the assay is not "they checked your gut and it was fine." It is "the assay that earned the word measured the things it was built to measure, and the gut-microbiome question was not one of them" — which is a known edge of what a young science had learned to ask, not a danger hiding behind a clean result.

And you do not have to take that from a finger run down an endpoint list. The regulator says it itself.

The regulator's own words

EFSA — the European Food Safety Authority, whose evaluations the GB system inherited — published new guidance in 2026 on exactly what an applicant must submit to get an additive authorised. Inside that guidance is a sentence that does more to settle this report than any study could.

The science, EFSA writes, "is not mature enough to include specific recommendations" on "the impact that the food additive may have on the gut microbiota composition and function (microbiome)." And the work to fold the microbiome into the safety framework is something whose "recommendations… are being developed."⁵

Read that twice, because both halves are load-bearing and they point in opposite directions and both are true. "Not mature enough to include specific recommendations" — that is the regulator telling you the endpoint is not yet a required part of the assay. "Are being developed" — that is the same regulator, in the same document, telling you it is actively building toward including it. The only place the gut bacteria enter the current required assay at all is as a metaboliser — a thing that might chemically alter the additive — never as a thing that might itself be affected.⁵

So when this report says the gut-microbiome question was not in the battery, it is not an accusation that anyone looked away. It is the institution's own current account of where the frontier sits. The microbiome was not a tractable safety-science object when these emulsifiers were first authorised — the emulsifier and stabiliser class has sat in the authorised framework since the mid-1970s, decades before the microbiome existed as a research field.⁶ A question cannot be required before it is askable. The labs that did eventually ask it — and we are about to meet them — asked it on their own research-grant logic, from outside the approval machine, because that is where new questions come from.

The applicant funds and submits the dossier. In EFSA's own words, the applicant "is responsible for providing all the available… scientific data that are pertinent to the safety assessment."⁵ A company compiling a submission funds the endpoints the regulation requires. It does not fund speculative, expensive, decade-horizon questions no rule demands and no field has yet learned to pose. That is not a charge against anyone's good faith. It is the simple geometry of who pays for which question — and the chronic-gut endpoint fell in the space between every actor's incentive, owned by none of them, until academic curiosity picked it up. The regulator still independently evaluates whatever is submitted; this is not self-certification. It is that the menu of required questions is written by what the science of the day knew how to ask.

The science that finally asked

The question got asked, properly, for the first time in 2015 — in mice.

A study in Nature found that two common emulsifiers, carboxymethylcellulose and polysorbate-80, given at low concentrations, drove the gut bacteria of mice to encroach toward the gut wall, with low-grade inflammation and features of metabolic syndrome; transplant experiments showed the microbiota changes were doing the work.⁷ It was a striking result. It was also entirely in rodents — animal-grade evidence, the question posed for the first time, not an answer about people.

The first careful human test came in 2022: the FRESH trial, an inpatient controlled-feeding study at the University of Pennsylvania. Sixteen healthy adults ate either an emulsifier-free diet or the same diet enriched with 15 grams a day of CMC, for eleven days.⁸ The CMC group's microbiome diversity fell; their gut metabolome was depleted. And in the finding that gets quoted most — and most needs its full shape kept intact — two of the seven people eating CMC showed bacteria moving into the normally bacteria-free inner mucus layer, while the other five did not, and the controls did not. The authors are explicit about what that is: a two-of-seven result that "yielded a 2-tailed P value of 0.175, which does not meet common standards of being statistically significant" — a "reasonable" signal, in their words, not a proven effect.⁸ Over those eleven days there was no significant change in faecal lipocalin-2, a marker of inflammation (the authors report P = .258).⁸ And the dose, the authors note, at 15 grams a day "likely exceeds CMC intake of most individuals."⁸

Then came the largest, most comprehensive human test to date — and it is the reason this report stays calm, so it belongs in the centre of the page, not in a footnote.

In 2025, a placebo-controlled randomised trial put sixty healthy adults on an emulsifier-free diet, then added — through brownies — carboxymethylcellulose, polysorbate-80, carrageenan, soy lecithin, native rice starch, or nothing, for four weeks.⁹ The microbiome moved: short-chain fatty acids fell in the people eating CMC, mirrored by other emulsifiers. But on the endpoints a worried shopper actually cares about, the trial's conclusion was flat: "Emulsifier supplementation did not impact intestinal or systemic inflammation or metabolic endpoints." No difference in faecal calprotectin, C-reactive protein, lipopolysaccharide-binding protein, cholesterol, or other metabolic markers.⁹

It is not a perfectly clean null, and honesty requires the wrinkle: in the people eating carrageenan — the weakest-evidenced of the three, the one with its own regulatory question mark — transcellular intestinal permeability rose, at P = .04, measured against their own baseline rather than against placebo, in what the authors describe as an exploratory trial.⁹ One exploratory blip in the most contested additive, against baseline, is not a harm signal. It is exactly the kind of result that holds open a young question without closing it toward danger.

Here is the discipline the rest of this report rests on, and the place I refuse to do the easy, vivid thing: the FRESH non-significant two-of-seven and the n=60 null are placed side by side and not reconciled into a single verdict — because no single verdict is honest. The microbiome shifts are reproducible. Clinical harm in healthy adults is not demonstrated. The response varies markedly between individuals — some people's microbiomes move, others barely register it. That three-part answer is genuinely what the evidence says. And it is precisely the answer a tidy code cannot carry, because a code on a shelf can only encode a verdict, and the truth here is not a verdict. It is a probability, an individual variation, and an open edge.

One more boundary, because the short list matters and the grades inside it are not equal. Carboxymethylcellulose is the only one of these emulsifiers with a human controlled-feeding trial behind it. Polysorbate-80 shares the mouse and the laboratory evidence but has no human keystone trial of its own; it was tested in the n=60 trial and showed no inflammation or metabolic harm.⁹ Carrageenan is the thinnest and most contested case — EFSA itself declared its ADI "temporary" in 2018, saying the database "should be improved within 5 years," and most of the alarming carrageenan literature concerns degraded carrageenan, or poligeenan, which "has not been authorised as a food additive and is not used in any food applications."¹⁰ Three additives, three different evidence grades. The rest of the additive universe is not implicated by any of this.

The same word, read two opposite ways

Now the behavioural turn, which is the actual subject of this report.

A shopper trusts the regulator. She sees "approved," reads "safe," and stops asking. A different shopper distrusts industry. She sees "E466," reads "synthetic, processed, suspect," and recoils. These look like opposite people. They are doing the identical thing. Neither has read the study. Both have read the code.

The fearful pole is documented and, importantly, traceable to its source. A 2019 study of food experts found that consumer distrust of E-numbers "arose from negative communication by traditional media, social media and books" — not from toxicology.¹¹ That fear is not coming from the evidence. It is coming from the orderly, scientific look of the code being read as a verdict of danger.

The trusting pole runs on the same machinery in the opposite direction. The perception literature finds that when people lack knowledge of the actual science, their risk and benefit judgements are driven by "their knowledge of regulation, their trust in regulators," and that trust matters most "in the absence of sufficient knowledge."¹² The less you know, the more the word does. Trust substitutes for the knowledge, and — this is the part that makes the mechanism so slippery — the substitution leaves no trace. Delegating a question to a trusted word feels exactly like having the answer. Which is why, if you ask the trusting shopper what, specifically, she is trusting when she reads "approved," she cannot say. Not because she is careless. Because the word was built to let her not have to.

This is the engine, and it is also the safety rail. If this report frightened the trusting reader — if it took the two-of-seven and the mouse study and made your bread feel like a threat — it would simply have flipped you to the other pole, the fearful one, which is just as uncoupled from the evidence. The whole finding is that both readings are wrong in the same way. There is no version of this report that scares you and is also right.

This is a different mechanism from the ones YAN has mapped before, and the difference is the whole point. In "The Permission Slip Economy" (031), the consumer makes a trade she is aware of — a virtuous purchase buys a little permission for something else; there is a ledger, and she can be shown it. In "The Detox Label" (091), the trust is placed in a marketing claim a brand is making. Here there is no trade and no ledger and no marketing — the trust is placed in an official, regulatory signal, the most authoritative pole there is, and the substitution erases its own trace so completely that the shopper cannot name what she handed over. You cannot make the ledger visible, because there isn't one. The only thing you can make visible is the scope of the word.

The word is a dated photograph — and the camera is still working

"Approved" is a snapshot of what its era knew how to ask. That cuts in a reassuring direction as often as a worrying one, and the system genuinely updates.

The clearest case is titanium dioxide, E171 — a long-permitted whitener. When EFSA re-examined it and concluded in 2021 that a genotoxic concern "could not be ruled out," the EU removed it as a food additive; from 7 August 2022 it was no longer permitted.¹³ The word was withdrawn the moment the evidence warranted. The machine works.

It also lags at the border, and that matters for a shopper in Reading. The same E171 ban applies in Northern Ireland, which follows the EU rule — but goods authorised in Great Britain containing titanium dioxide can still be moved into Northern Ireland under the retail movement scheme, because E171 remains a permitted food additive in Great Britain.¹³ The same molecule: "no longer safe" on one side of a line, "permitted" on the other. "Permitted in GB," then, does not mean "endorsed by the most recent science everywhere." It means "still on the GB list, as of the last time the GB list was revised."

So the snapshot is dated, and it is being re-shot — slowly, unevenly, sometimes stopping at a jurisdictional border. The 2026 EFSA guidance beginning to name the microbiome is the camera turning, finally, toward the one question this whole report is about.⁵ The reassurance is largely warranted. The list does update. And the specific question Sarah might care about is only now entering the queue of things the system knows how to ask.

The public is already naming the gap

In April 2026, UK Research and Innovation commissioned a public dialogue — 132 members of the public, deliberating, facilitated independently — on ultra-processed food. The participants did not need a behavioural framework to find the gap. They described food marketing as "lulling people into a false sense of security," said "choice is an illusion," and reported feeling, in the project's own headline, "left in the dark."¹⁴

That is the substitution, named from the inside by the people experiencing it. The unease is not about a proven harm. It is about not being able to tell — from the pack, from the word, from the shelf — what has been studied and what has not. The public has put its finger on the exact epistemic gap. The thing it cannot find is the scope of the word.

Two cautions on that, both honest. The dialogue was about ultra-processed food broadly — marketing, advertising, obesity — not specifically about additive endpoints; it supports "the public is naming an epistemic gap," not "the public has confirmed additives harm the gut." And a 2026 British Medical Association report concluding that the UK's "current approach to regulating, producing, marketing, and distributing food is failing to protect public health" is likewise a broad ultra-processed-food argument, not an additive-endpoint finding.¹⁵ The gap these voices name is real and it is epistemic. It is not a verdict on the chemistry.

The gap, stated plainly

Here, finally, is the whole of it, and it is narrower than either pole wants it to be.

"Approved" certifies that an additive passed a prescribed, applicant-funded toxicology battery and earned a place on a list, and has not been re-evaluated off it — a real, meaningful safety signal. But the same tidy code that earns the word converts an unasked, narrow, still-open question into felt certainty in whichever direction the shopper's priors point. So the word does not add safety knowledge to her mind. It removes the felt need for it.

The residual gap is not poison in the trolley. It is not a regulator who failed — EFSA is, in its own words, developing the framework. It is one millimetre wide and it sits on the shelf: you cannot tell, from the pack, which of your additives carries a live scientific question and which is settled. The loaf does not say "the emulsifier in here is one of three with an open microbiome question; the other twelve ingredients are not." No label encodes the scope. That is the gap. It is real, it is small, and it survives no matter how the emulsifier science finally resolves — because it is a gap in the word, not in the body.

I should name the strongest case against even that narrow claim, because it is a good one and the evidence makes it. The orderly code is a genuine cognitive economy — a shopper cannot re-derive the toxicology of every ingredient at the shelf and should not have to; the alternative to trusting the list is paralysis or panic. And the best human evidence to date is reassuring: the largest trial found no inflammation and no metabolic harm, the human keystone's most alarming result is statistically non-significant, the one positive permeability signal is in the weakest additive against a baseline. A fair, well-disposed person could finish here and conclude: the system works, the science is simply young, and there is nothing to see. Most of that, I think, is right — which is why the residual gap I am claiming is not "the system failed" but only "the word on the shelf carries none of this, so you cannot tell which question is open." That smaller claim survives even if every reassuring thing in the paragraph above is true.

What would change this picture? A larger, longer human trial at habitual intake levels showing real clinical inflammation, permeability damage, or metabolic harm from CMC or polysorbate-80 in healthy adults would move the gap from "unstudied endpoint" toward "evidenced harm" — and then the calm register of this report would be wrong, and the worried shopper's instinct right. Run it the other way: an EFSA re-evaluation concluding the microbiome shifts carry no health consequence at dietary intakes would narrow the gap further toward "the shelf doesn't flag a low-stakes individual finding," and tilt the honest answer harder toward "carry on." And the behavioural claim at the centre of this — that trust in a regulator substitutes for the felt need to know — rests on perception research, including a single cross-sectional study whose inverse-knowledge direction has not been independently replicated at a stated effect size;¹² if that fails to replicate, the substitution mechanism is real but smaller than this report implies. Each of those is a real, obtainable result. None of them is in hand yet. That is the honest state of the question.

What to do with the word

Most of this resolves to one piece of literacy, and then almost nothing.

Decode the word once, and keep the decoding. "Permitted," or an E-number, means not acutely toxic at expected intake and doing a technological job — not "studied for your gut over years and cleared." That single line corrects both poles at once: the panic that reads the code as poison, and the trust that reads it as a personal long-term guarantee. Knowing what the word is for is the entire literacy payload. It is a way to think about the word once, not a new thing to monitor.

Know the short list, not the alphabet. The two or three additives with a genuine emerging human-and-mechanistic evidence base are carboxymethylcellulose (E466) and polysorbate-80 (E433), with carrageenan (E407) as the weaker, contested case — and even there the strongest human finding is "the microbiome shifts; clinical harm is not demonstrated." The rest of the additive universe is not implicated. Recognising those two or three is the whole of it. The fear that treats every E-number as suspect is the error this exists to correct, not encourage.

And then, for most people, most of the time: do nothing differently. That is not a hedge or a cop-out. It is the evidence-warranted read — the largest trial null on the endpoints that matter, the intakes within their thresholds, the harm hypothesis one contested strand of a broader debate rather than an established fact. Saying so plainly is the honest answer, not a failure of nerve.

If you want a low-stakes lever anyway, here it is, framed as exactly what it is. Where the swap is free or cheap — and only there — a shorter-ingredient loaf, the kind with a flour-water-yeast-salt-led list and no emulsifier, is a real and widely available category in any mainstream supermarket's own-label range. The honest trade-off is not "poison versus safe." It is that emulsifier-free bread tends to stale faster and sometimes costs more, because the emulsifier is a crumb-softener and shelf-life extender doing a real job. That is the actual cost of the swap.

No specific brand is named here, on purpose, because the honest ending is the one the shelf forces. No label currently tells you which of your additives carries a live scientific question and which is settled. For that to change, a label — or the FSA's own consumer page — would have to add the one thing it omits: the scope of the word. Not "safe," but "safe for what, tested for how long, and which of these is mid-re-evaluation." Until a signal like that exists, the literacy lives in your head, not on the pack: the next time you read "approved" on an ingredients list, you will know it is a real safety signal answering a narrower question than the one you were silently asking it — and that knowing which question is which is, for now, the one thing the word will not do for you.