FDA holds 1,325 food substance notices and no count of the rest — the rule to require notice is still unproposed after 16 years

What is happening?

A company that wants to put a new substance into food sold in the United States may conclude on its own that the substance is generally recognized as safe, market it, and never tell the Food and Drug Administration. As of 2026-08-04 the public FDA inventory of substances it has been told about holds 1,325 notices. There is no inventory of the substances it has not been told about, and no agency publishes a count of them.

The operative regulation is 21 CFR 170.205, adopted by the FDA final rule at 81 FR 54960 on 2016-08-17. Its title is the word Opportunity, not the word Requirement. Under it any person may notify FDA of a view that a substance falls outside premarket approval. No companion section requires anything of a person who declines the opportunity.

The carve-out underneath is statutory rather than discretionary. Section 201(s) of the Federal Food, Drug, and Cosmetic Act excludes generally recognized as safe uses from the definition of food additive, and section 409 imposes premarket approval only on food additives. FDA describes its own regime in the current regulatory agenda in the same terms — a person who reaches such a conclusion may, but is not required to, notify FDA, and the submission is therefore currently voluntary.

In March 2025 the Secretary of Health and Human Services directed FDA to explore a rulemaking that would end the pathway. That rulemaking carries the regulation identifier RIN 0910-AJ02. As of 2026-08-08 the Federal Register contains zero documents under that identifier. Its target publication month has moved from October 2025 in the Spring 2025 regulatory agenda to December 2026 in the 2026 agenda, and 22 meetings under Executive Order 12866 had been logged on it as of 2026-08-07, on a rule that does not yet exist as a public document.

Whose problem is this?

The body that would have to write the rule stated in 2016 that it lacks express statutory authority to require the notices. The body that could supply that authority did supply it once, for food contact substances, by statute, in 1997. No statute closing this pathway has been enacted, and nothing in the record obliges either body to move.

Where does this problem end?

The boundary matters here because the exemption is claimed by the firm rather than granted by the agency. A claim that requires no filing leaves no record behind it, so the size of the exempt population is unknown by design and not by oversight.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names a number of substances to be brought under notification, a share of the food supply to be covered, a date by which coverage would be complete, or a measure by which anyone would later judge whether the rule worked. The 2026 agenda entry prints Risks as TBD and Alternatives as TBD, and prints the legal deadline as none.

How big is it?

Between 336.8 million and 341.8 million people, which is to say everyone eating food manufactured and sold in the United States. The US resident population was 336,755,052 in 2023 and 341,784,857 in 2025 on the same World Bank series, and those two vintages are the low and high ends of the band carried in the population file.

That band should be read for what it is. Its width is the spread between two vintages of one population estimate, not a range of exposure uncertainty. It is narrow because the honest answer to how many people are exposed is everyone who eats from the commercial food supply, and no source opened here supports subsetting that. A wider and more informative band cannot be built, and the reason it cannot is the problem itself — subsetting would require knowing which substances rode the un-notified pathway and which products carry them.

Ultra-processed foods supplied a mean 55.0 percent of total calories among US residents aged 1 and older between August 2021 and August 2023, with 61.9 percent among youth aged 1 to 18 and 53.0 percent among adults aged 19 and older. That figure is carried here only to show that industrially formulated food is the majority calorie path. It is a calorie share and not a headcount, its source never mentions this pathway, and nothing published supports using it to size a subgroup.

The number counts exposure and not harm. No source opened here links any specific un-notified substance to any health outcome in the United States.

Under what conditions does it arise?

1. The exemption is claimed, not granted. Section 201(s) removes generally recognized as safe uses from the definition of food additive, and section 409 reaches only food additives. A firm marketing on its own conclusion is doing what the statute permits rather than evading a rule. Nothing conditions the carve-out on registering, disclosing or being reviewed.

2. The agency conceded the missing instrument in writing. In Response 28 of the 2016 final rule preamble FDA agreed that it lacks express statutory authority to require companies to submit notices, and set out the contrast — Congress amended section 409 in 1997 to establish a mandatory notification program for food contact substances, and has never amended it to establish one here. The tool exists and has been used. It was simply never pointed at this pathway.

3. The experts who decide are selected and paid by the firm. A peer-reviewed study published 2023-09-06 examined 403 notices filed between 2015 and 2020 and found 57 percent relied on an expert panel, 30 percent on employees of the notifier and 13 percent on employees of a third party, and that seven individuals held 339 of 732 panel seats, which is 46 percent. Those are the panels behind notices FDA sees. The same market of panelists serves conclusions FDA is never told about, with no filing, no letter and no public record of who sat on the panel.

4. Exit costs far more than entry. For a substance already in the food supply the only vehicle is 21 CFR 170.38, under which FDA must find that consensus among qualified experts has ceased to exist. That is a negative to be proved rather than a hazard to be shown, and the one comparable case ran from a citizen petition in 2004 to a compliance date of 2018-06-18.

A firm can price a paid expert panel to the dollar before it ships. The public cannot price what it does not know is there.

5. Nothing sets a clock. The regulatory agenda entry prints the legal deadline as none. There is no statutory trigger, no consent decree and no reporting duty whose breach anyone could point at.

What has been tried?

Two directions have been tried. One asks the agency to require the notices, and it has run 16 years from the 2010 recommendation without a proposed rule. The other removes a substance after it is already in food, and the block below gives the reason it is used approximately never.

What was found?

Why is it still unsolved?

Institutional exemption — the conduct is carved out of the requirement by statute, and an exemption a firm claims for itself leaves nothing behind to count.

The ordering is what fixes the type. The missing inventory is downstream of the carve-out, not beside it. Close the measurement gap alone and you have a list of substances FDA still could not review, because section 201(s) would still put them outside premarket approval. Close the carve-out and the list follows automatically, which is why the FDA proposal runs in that second direction and treats a public inventory as a byproduct. The conduct here is not unlawful and not unmeasured by accident. It is permitted, and the permission is self-asserting.

The second movement is that the actor with the incentive lacks the clean instrument, and the actor with the instrument has not used it. FDA wrote in 2016 that it lacks express statutory authority to compel the notices, and that mandating them would require a separate rulemaking. That separate rulemaking did not begin for nine years. Its proposed device is a presumption that the exemption does not apply unless a notice has been filed, which turns an exemption a firm asserts into one it must register to claim. Whether an agency may do that without the kind of statutory amendment Congress supplied for food contact substances in 1997 is the question the rule will be litigated on, and it is a plausible reading of why the draft has been at the review office long enough to host 22 recorded outside meetings.

The third movement is that the delay is produced by the structure rather than by anybody in it. A rule classified economically significant has to quantify its costs on a population — the substances nobody notified — that no one has counted, because the exemption made counting unnecessary. So the rule must either estimate the very thing whose absence motivates it, or proceed with its risks and alternatives printed as TBD. It has done the latter, and between two published abstracts its scope visibly narrowed. Nobody has to act in bad faith for this to take a decade. The exemption removes the information the rulemaking needs in order to justify removing the exemption.

Underneath all three sits an asymmetry of proof. To bring a substance in, a firm needs a favorable conclusion from experts it selects and pays, and no filing at all. To push one out, FDA must run 21 CFR 170.38 and establish that expert consensus has ceased to exist. The partially hydrogenated oils case shows the price of that direction — 14 years from petition to compliance, against an FDA estimate of USD 130 billion in net benefits from the removal.

What observation would mean it is solved?

Candidates — (a) FDA publishes a final rule requiring notification, with a stated compliance date (b) the public inventory of notices rises sharply and then levels off, indicating the previously un-notified population has come in (c) an independent recount of substances marketed on an un-notified conclusion is published, and it falls.

(a) alone is weaker than it looks. A final rule can arrive with a scope narrow enough that the population it was written for stays outside it. The 2026 abstract already reads certain uses without saying which, and it no longer carries the indirect-substance language or the two exemptions the Spring 2025 abstract named. A rule can also be vacated in litigation, and the previous rule in this area was challenged the year after it issued.

(b) alone measures the wrong population. A rising notice count records compliance by firms that were going to comply. It cannot see a firm that keeps its conclusion to itself, which is precisely the group in question. The same inventory omits the filing date for 647 of its 1,325 records, so an arrival rate cannot even be computed from FDA data for roughly the last decade, and closure dates are a different quantity.

(c) alone has no baseline. The only published estimate is about 1,000, made in 2011 and restated without a recount since. A new count would be the first measurement rather than the second, so it could not show a fall. The three have to be read together, and (c) has to be read as establishing a baseline before it can measure anything.

What is it connected to?

Fills with researchfood additive regulation in other jurisdictions, state-level food chemical laws inside the United States, the regulatory review process at the office where this draft currently sits, and the relationship between the mandatory food contact notification program created in 1997 and this pathway. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• How many substances are in the food supply on an un-notified conclusion. Not the 2016 final rule preamble, which discusses the question at length and answers it with legal authority rather than a number. Not the 2025 information collection notices. Not the regulatory agenda entry, which is the sharpest case — that entry is rated economically significant and defines the recurring cost of the rule as the cost of notifying for substances that would otherwise have been the subject of an independent conclusion. That cost cannot be priced without a count, and no count is published anywhere.
• What the rule is meant to achieve. The agenda entry prints Risks as TBD and Alternatives as TBD for an economically significant rule that has already hosted 22 recorded outside meetings.
• Why the FDA burden table does not add up. 100 plus 12 is printed as a total of 125, in the 60-day notice of 2025-06-27 and again in the 30-day submission of 2025-08-26. It reached the review office unreconciled and no correction was printed.
• Why the public inventory omits the filing date for 647 of 1,325 records. Every record from GRN 614 upward lacks it, so notices filed per year cannot be computed from FDA data for roughly the last decade. No note on the page mentions the omission or explains it.
• What the Substances Added to Food inventory is partial of. FDA states on the face of the list that it is only partial, names the categories it excludes, and publishes no count for any of them. The denominator behind 3,971 is unstated.
• What to call the pathway. Federal Register full-text searches run 2026-08-08 return 5 documents for the industry and press term self-affirmed GRAS, all of them EPA pesticide documents and none from FDA on food; 1 document for the phrase FDA used in the 2016 rule; 2 for a second FDA phrasing; and 6 for a third. The agency has written about this conduct in at least four phrasings across three decades and settled on none of them, so there is no term of art a regulation could be searched for.
• Who is meeting the review office. The public record gives date, agency, rule title, stage and meeting type for the 22 meetings. Per-meeting attendee detail was not reachable in this round, so the number of contacts on an unpublished rule is public while the identity of the participants, at that surface, is not.
• Why the one open docket drew nothing. The 60-day information collection notice received zero comments, in the same year the pathway became a national political issue with a Secretary-level directive behind it. No source opened here remarks on the gap between the volume of public argument and the volume of public comment.

See the evidence

The primary documents in this file were read directly. The 2016 final rule, the two 2025 information collection notices and the 2015 final determination were retrieved as Federal Register full text and searched locally, so the regulatory text, the burden table and the cost figures are first hand. The two FDA inventories were opened and the notice inventory was downloaded in full and tabulated here — the shares of 79.7, 17.0, 1.5 and 1.8 percent, the 647 records without a filing date and the closure counts by year are all computation over that download and are printed by no source. The two agenda entries and the meeting search were opened directly, and the absence of any Federal Register document under the identifier was established by query rather than by assertion. Secondhand material is confined to four places — the panel concentration study, the two trade articles, and the population and calorie figures. Two disagreements are left visible rather than resolved. FDA prints 100 plus 12 as a total of 125 in two consecutive notices, so 112 and 125 are carried as an unreconciled pair and are not averaged. And the figure of about 1,000 un-notified substances has a lineage rather than a source — one trade article attributes it to a 2013 estimate and misnames the organization that produced the underlying work, a 2023 peer-reviewed paper restates it as prior work rather than its own count, and the 2011 article beneath both could not be opened. It is presented throughout as an estimate made in 2011 and restated since, never as a current count. The counts in this document also have different denominators and are deliberately not cross-checked against one another — 1,325 notices filed since 1998, 1,056 cleared, 100 or 112 or 125 estimated per year, and 73 closures in 2025 answer different questions, and reconciling them would manufacture a number no source states. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 336,755,052–341,784,857 As of 2023-2025

Sensitivity The width of this band is the gap between the 2023 and 2025 vintages of one population series and says nothing about how uncertain exposure is. The band counts exposure and not harm, because no source opened in this round links any specific un-notified substance to any health outcome in the United States. It also has no numerator: the natural framing of N people exposed to M unreviewed substances cannot be completed, because M is exactly the count the missing inventory would supply, and the uncountability of M is the problem itself rather than a gap in the research. It cannot separate reviewed from unreviewed exposure either, since nothing on a label distinguishes a substance among the 1,056 that FDA replied it had no questions about from one FDA was never told of. In the opposite direction the limit is coverage rather than value: the band excludes people outside the United States who eat food manufactured in the United States on this basis, and it does not account for imported foods whose ingredients entered US commerce through the same exemption.

Regional breakdown No source opened in this round reports substances marketed on an un-notified GRAS conclusion, or the products containing them, at any geographic level. A state or regional value would require the substance list that does not exist, and splitting the national figure by population share is not permitted.