The Singapore Food Agency published its first List of Approved Novel Foods on 17 March 2026 — 23 pages, 14 products, every decision it has taken between 21 May 2019 and 17 October 2025. For six years the only public record of what Singapore had cleared was a patchwork of company press releases and trade coverage. There is now a primary document.
It is worth reading for three reasons, in ascending order of usefulness. It settles some dates that the industry has been citing loosely. It discloses process and specification detail that no applicant ever published. And it establishes what a regulator is willing to put in public when it decides to publish at all — which matters to every other jurisdiction now drafting a framework.
The first decision it records, on 26 November 2020, is the cultured chicken clearance universally described as the world’s first approval of cultivated meat. It is normally dated to December 2020.
The full list, with the dates SFA records
Entries appear here in SFA’s own order. Attributions in the right-hand column are from trade reporting, not from the file — SFA does not name applicants, though three products appear under their trade names in the process descriptions.
| # | Novel food | SFA decision date | Key published specification | Attributed to (trade press) |
|---|---|---|---|---|
| 1 | Chlamydomonas reinhardtii (green) strain THN6 algae biomass | 21 May 2019 | crude protein 30–70%, moisture ≤10% | Triton Algae Innovations |
| 2 | Cultured chicken cells (UMNSAH/DF1) | 26 November 2020 | protein 5–10%, moisture 85–95%; media DMEM/F12 with FBS | Eat Just / GOOD Meat |
| 3 | Mycelial biomass, Fusarium strain flavolapis | 6 August 2021 | protein >45%, water 70–80%, RNA under 2.0% | Nature’s Fynd (Fy protein) |
| 4 | Protein powder from Xanthobacter sp. SOF1 (“Solein”) | 29 September 2022 | protein ≥73% dry weight, endotoxin ≤540 EU/g | Solar Foods |
| 5 | Chlamydomonas reinhardtii (red) strain TAI114 | 19 October 2022 | crude protein 30–70%, protoporphyrin IX 3–7% | Triton Algae Innovations |
| 6 | Cultured chicken cells, serum-free process | 11 January 2023 | protein 5–10%, moisture 85–95% | Eat Just / GOOD Meat |
| 7 | Mycoprotein from Neurospora crassa Bstr 26 (“Rhiza”) | 15 October 2024 | protein >40%, ash under 10% | The Better Meat Co |
| 8 | “Fermotein”, Rhizomucor pusillus fungal biomass | 14 March 2024 | crude protein 44–68%, fibre 25–39%, moisture 3–10% | The Protein Brewery |
| 9 | Cultured Japanese quail fibroblast | 22 March 2024 | protein >4%, moisture >80%; bioreactors 200–20,000 L; harvest at minimum 30 g/L | Vow |
| 10 | Urolithin A from chemical synthesis | 13 March 2025 | assay ≥98% HPLC area | not identified |
| 11 | 3’-sialyllactose sodium salt (E. coli BL21(DE3) JBT-3SL) | 18 June 2025 | 3’-SL ≥88% dry weight | not identified |
| 12 | 6’-sialyllactose sodium salt (E. coli BL21(DE3) JBT-6SL) | 18 June 2025 | 6’-SL ≥90% dry weight | not identified |
| 13 | Mycelium biomass from Pleurotus pulmonarius | 24 September 2025 | protein ≥25 g/100g dry weight | not identified |
| 14 | Cultivated chicken biomass (Gallus gallus embryonic stem cells) | 17 October 2025 | protein >6%, moisture >70%; 12 numbered critical control points | Parima (Gourmey/Vital Meat) |
Two structural notes. The list is not strictly chronological — entry 7 (October 2024) sits ahead of entries 8 and 9 (March 2024). And four of the fourteen are not alternative protein at all: Urolithin A is a chemically synthesised compound and two entries are human milk oligosaccharides. Singapore’s novel food framework is being used as a general pre-market gate, not a cultivated-meat gate, which is a useful corrective to how the regime is usually described.
Three dates that do not match the ones in circulation
26 November 2020, not December 2020. The clearance of cultured chicken cells derived from the UMNSAH/DF1 line — the world-first — is dated 26 November 2020 in SFA’s record. The December date is so embedded that Keller and Heckman’s own note on this very document says SFA “granted the world’s first approval for cultivated meat in December 2020”. The gap is a month, and it is the difference between a decision date and an announcement date.
22 March 2024, not April 2024. The cultured Japanese quail fibroblast approval — reported at the time, and again in April 2026 coverage of this list, as an April 2024 approval — is recorded as 22 March 2024.
Fermotein preceded Rhiza. Coverage of the list describes two 2024 mycoprotein approvals with Rhiza first and Fermotein second. SFA’s dates put Fermotein on 14 March 2024 and Rhiza on 15 October 2024, seven months apart in the other order.
None of these are consequential on their own. Together they make a point that recurs across this sector: the public timeline of regulatory events is assembled from announcements, and announcements are marketing artefacts scheduled for convenience. A regulator’s own decision date is a different data type. Anyone computing submission-to-clearance intervals — as we did when counting twelve FDA clearances for precision-fermented protein — should use the register, not the press release, and should expect the two to differ by weeks.
Practical consequence. If your regulatory timeline model is built on trade-press dates, it is biased late by an amount you cannot estimate. Rebuild it from registers where registers exist: SFA’s list, the EU Union list, the FDA GRAS inventory.
What the file discloses that the companies did not
This is the part that repays the download.
A harvest density and a bioreactor range for cultured quail. Entry 9 states that cells are expanded “in a stirred tank bioreactors ranging from 200 – 20,000L under aseptic conditions” and harvested “upon reaching the target culture volume and density (minimum of 30 g/L)”. A minimum harvest density is the single most load-bearing number in a cultivated-meat cost model, and it is almost never published. Thirty grams per litre is a serious figure — it is in the same range as the titres Tetra Pak used in its own bioreactor batch-loss model — and it comes with a vessel range that tells you the scale the process was assessed at.
Fetal bovine serum, on the record, in the 2020 file. Entry 2 describes growth media “supplemented with FBS as described in the applicant’s submission”. Entry 6, the January 2023 serum-free version of the same cell line, describes the same basal media without it. The transition that the sector narrated in press releases is legible here as two regulatory files 26 months apart.
A medium composition for the 2025 cultivated chicken biomass. Entry 14 lists “sugars, vitamins, amino acids, proteins, minerals, polyamines, vegetable fat and anti-shearing agent”, states the cell line carries no genetic modification, and enumerates twelve critical control points — cell bank qualification, cleaning-in-place, sterilisation, sterilising-grade filtration of the medium, filter integrity checks, visual inspections, allergen testing on regulatory batches, and storage at −20 °C. That is closer to a process outline than anything the company has published.
Specification ranges wide enough to matter to a formulator. Fermotein’s crude protein is specified at 44–68%. Its maker describes the ingredient publicly as “about 50% complete protein, 35% healthy fibers”. Both are true — one is a regulatory range, the other a typical value — but a buyer contracting on “50% protein” and a supplier shipping against a 44–68% specification are not agreeing on the same thing. This is the measurement problem we found in crude versus true protein and in capacity tonnes with no stated basis, showing up here in the regulatory layer.
Cultivated products are 4–10% protein, wet. Entries 2, 6, 9 and 14 specify protein at 5–10%, 5–10%, >4% and >6%, with moisture between 70% and 95%. Every cost-per-kilogram figure quoted for cultivated meat refers to material of roughly this composition. It is worth holding beside the four cost figures for cultivated duck that shared no basis.
What it withholds, and why that is a deliberate design
SFA does not publish applicant names, and does not publish the dossiers. Keller and Heckman read this as maintaining “some level of confidentiality, while offering transparency on the products themselves”, which is the right reading — it is a considered trade-off rather than an oversight.
The cost of that choice is that the list cannot be used for the thing most people want it for: tracking which company is cleared where. Three products are identifiable anyway because their trade names — Solein, Fermotein, Rhiza — appear inside the process descriptions, and most of the rest are identifiable by strain from public announcements. Two are not identifiable from public information at all: the Pleurotus pulmonarius mycelium cleared in September 2025 and the synthesised Urolithin A.
Practical consequence. For competitive intelligence the list is weak. For technical work it is strong — if you are preparing a dossier for a similar organism, entries 3, 7, 8 and 13 tell you the specification structure SFA accepted for four different fungal biomasses, including which parameters it required (RNA content for one, endotoxin for a bacterial biomass, heavy metals and a defined pathogen panel for all).
The counter-argument
A published list is not a published assessment. SFA gives no reasoning, no intake limits, no conditions of use and no scientific opinion — unlike EFSA, which publishes a full opinion, or the FDA, which publishes the notice and its response letter. A reader can see that something was cleared and to what specification, but not why, or on what evidence.
That is a real limitation, and it means the list should not be read as evidence that Singapore’s standard is equivalent to, or lighter than, anyone else’s. It is also worth noting the political context: Singapore’s environment minister has said alternative proteins are no longer part of the country’s food strategy, citing higher production costs and weaker-than-expected consumer acceptance. Publishing a six-year back-catalogue is compatible with a regime that has become less enthusiastic, not more.
What we could not establish
- Whether SFA will keep the list current. The revision history contains one line: “17 March 2026 – Initial Issue.” No update cadence is stated.
- The applicants behind entries 10 and 13. Neither the file nor the coverage we read identifies them.
- Whether decisions after 17 October 2025 will be added retroactively. At least one further cultivated approval was reported in the weeks after publication; the file as issued does not include it, and we did not find a stated policy for adding new entries.
- Whether “date that SFA/ex-AVA allowed the novel food to be used as food” is the decision date or the notification date. The column heading is precise about what it means but not about when the applicant learned of it, which is where the discrepancy with announcement dates most likely originates.
- Conditions of use. The file specifies composition, not permitted food categories or maximum levels. Whether such conditions exist and are communicated privately, we could not determine.
What to watch
- A second issue of the list. If SFA updates it — particularly if it adds the approvals granted since October 2025 — this becomes a maintained register rather than a one-off snapshot, and its value multiplies.
- Whether other regulators copy the format. A specification-and-process disclosure that omits the applicant is a genuinely new template, and it is one a jurisdiction nervous about commercial confidentiality could adopt quickly.
- Whether the 30 g/L figure gets used. It is now the only harvest density for a cleared cultivated product in any public regulatory record we are aware of. Cost models that currently assume a density should say whether they are above or below it.
- Whether specification ranges start appearing in supply contracts. A 44–68% protein range in a regulatory file is an invitation to argue about what “50% protein” means on a purchase order.