Industries / Food, farming and forestry / Food and beverage processing / Pet food
Safety and recalls
Safety and recalls is the work of keeping poisons, germs and wrong nutrient levels out of pet food, and of finding and pulling back bad batches when they get through.
7
Milestones achieved, from 2007 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2007
Melamine in imported gluten
In March 2007 cats and dogs across North America went into kidney failure after eating wet food from one contract maker that supplied many brands. Its wheat gluten from China held melamine and cyanuric acid, two industrial chemicals that together form crystals in the kidneys; of 451 cases that met a North American survey's criteria, 278 pets died, and the incident accounts for almost a quarter of the 3,691 dog and cat recalls the FDA listed from 2003 to 2022.
LessonDogs in Korea had died with the same kidney crystals in 2003 and 2004, in an outbreak blamed at the time on a mould toxin; Yhee and colleagues found melamine and cyanuric acid in the stored kidneys only after 2007.
Credit
Dobson and colleagues (Toxicological Sciences, 2008) identified the contaminants; Rumbeiha and colleagues ran the survey of laboratories and clinics
2008
Japan's pet food law
After the US deaths of 2007, Japan's agriculture and environment ministries set up a study group, and on 18 June 2008 they promulgated the Pet Food Safety Act for dog and cat food. It lets the government set limits, such as 2.5 µg/g (micrograms per gram) of melamine and 0.02 µg/g of aflatoxin B1, makes manufacturers and importers register and keep records, and requires labels to give the ingredients, a best-before date and the country of origin.
Credit
Japan's Ministry of Agriculture, Forestry and Fisheries (MAFF) and Ministry of the Environment
2012
Dry dog food sickens people
On 2 April 2012 Michigan's agriculture department found Salmonella in an unopened bag of Diamond brand dry dog food during routine shop testing, and the PulseNet laboratory network matched it to sick people. The CDC counted 49 people infected with the outbreak strain, 47 in 20 US states and 2 in Canada, linked to one Diamond Pet Foods plant in Gaston, South Carolina; 10 of the 24 with information were taken to hospital.
LessonThe CDC expected cases to continue for months after the recall, because dry pet food keeps for about a year and recalled bags stayed in homes.
Credit
Michigan Department of Agriculture and Rural Development; US Centers for Disease Control and Prevention (CDC); FDA
2018
Australia's unexplained dog illness
From late 2017 more than 100 dogs that had eaten Advance Dermocare dry food, made by Mars Petcare, were diagnosed with megaesophagus, a widening of the gullet that stops food reaching the stomach, and eight were put down. Mars recalled the food on 24 March 2018, but neither its tests for 160 possible toxic agents nor the University of Melbourne's investigation had found a cause when a Senate committee inquired that year and concluded that self-regulation of the industry "is no longer acceptable".
LessonIn November 2025 the consumer group CHOICE still described Australia's pet food standard, AS 5812:2017, as voluntary, with no authority to enforce it.
Credit
University of Melbourne's U-Vet Animal Hospital; the Senate Rural and Regional Affairs and Transport References Committee
2021
Aflatoxin kills dogs again
On 30 December 2020 Midwestern Pet Foods recalled Sportmix dry foods after reports of at least 28 dead dogs, and Missouri's agriculture department found very high levels of aflatoxin, a liver poison made by mould on corn. By 21 January 2021 the recall covered all corn-containing food from the company's Oklahoma plant, and the FDA knew of more than 110 dead and 210 sick pets, a count it called approximate and not all confirmed by laboratory tests.
LessonAflatoxin had already caused US pet food recalls between 1996 and 2008, and a review of those recalls named inadequate testing of raw materials among the causes.
Credit
Missouri Department of Agriculture; FDA
2022
Antifreeze in Brazilian treats
In 2022 dogs across Brazil fell ill or died after eating treats made by Bassar Pet Food. Minas Gerais state police found ethylene glycol, the poisonous main ingredient of antifreeze, in a treat, and Brazil's agriculture ministry (MAPA) traced it to two lots of propylene glycol, a permitted additive, bought from the supplier Tecno Clean Industrial; by 6 September more than 40 animals had died or fallen ill, and MAPA told every registered maker to find and recall products made with those lots.
Credit
Polícia Civil de Minas Gerais; MAPA
2025
Bird flu named a hazard
After H5N1 bird flu infected house cats in California, Colorado, Oregon and Washington that had eaten contaminated food, the FDA on 17 January 2025 required US makers of cat and dog food who use uncooked poultry or cattle materials, raw milk or raw eggs to add H5N1 to the written food safety plans that US law requires of them. Cooking, pasteurizing or canning inactivates the virus; raw food has no such step.
LessonIn September 2025 the FDA and the USDA's National Veterinary Services Laboratories linked a San Francisco cat's death to two lots of RAWR raw cat food, whose maker said it used only USDA-inspected, human-grade chicken; the virus belonged to a lineage last detected in November and December 2024, so it had survived in the frozen food for months.
Credit
FDA Center for Veterinary Medicine
Watch for
A faster Listeria screen
FDA scientists report a LAMP test, a DNA test that runs at one steady temperature, for Listeria monocytogenes that matched the FDA's standard culture method in raw pet food and soft cheese and gave reliable results at least 3 days sooner, in a validation run at one laboratory.
Achieved whena validation across several laboratories passes and the method is added to the FDA's Bacteriological Analytical Manual or used routinely by laboratories outside the FDA.
Claimed by
Beilei Ge and colleagues, FDA Center for Veterinary Medicine, September 2026
Open
01
One table of world recalls
A public, regularly updated dataset of pet food recalls and alerts from regulators in several countries, such as the FDA, the EU's Rapid Alert System for Food and Feed (RASFF), Britain's Food Standards Agency, the Canadian Food Inspection Agency and the Pet Food Industry Association of Australia's recall notices. Each row gives the date, the countries of manufacture and sale, the product type (dry, wet, raw, treat), the hazard in one shared code list, the lot or date range, illness and death counts with whose count they are, and the source; rows that share an ingredient lot or a plant across countries are linked.
The briefCredit · unclaimed02
Read owner posts for outbreaks
A test, on past events, of whether public posts by pet owners would have flagged a pet food outbreak before the recall. For each event, such as the aflatoxin recall of December 2020 or the Hill's vitamin D recall of January 2019, collect the public posts that name the brand from the months before, group them by brand and symptom with a classifier, and record the date an alert would first have fired against the date of the recall. Publish counts only, never user names.
The briefCredit · unclaimed
Open 01 · the brief
One table of world recalls
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, regularly updated dataset of pet food recalls and alerts from regulators in several countries, such as the FDA, the EU's Rapid Alert System for Food and Feed (RASFF), Britain's Food Standards Agency, the Canadian Food Inspection Agency and the Pet Food Industry Association of Australia's recall notices. Each row gives the date, the countries of manufacture and sale, the product type (dry, wet, raw, treat), the hazard in one shared code list, the lot or date range, illness and death counts with whose count they are, and the source; rows that share an ingredient lot or a plant across countries are linked.
Why it is still open
US recalls are well covered and nothing else is. DeBeer and colleagues counted 3,691 FDA dog and cat recalls from 2003 to 2022; the PerBowl dataset (CC BY 4.0, November 2026) lists 283 FDA-classified recalls since 2012 with lot codes; and the "pet food recall explorer" on GitHub tracks 331 FDA recall events since 2003. All three use FDA data only. Lorenzen and colleagues' RASNEX tool (PLOS ONE, 2021) mines RASFF notices for supply chains but is not built for pet food. Without a joined table, a question such as whether one bad ingredient lot caused recalls on several continents, as gluten from China did in 2007, cannot be answered from public data.
Built with what exists
The public RASFF Window portal; the FDA Enforcement Report, which can be filtered to its Veterinary product type; national recall pages; an AI model to pull fields from notices written in several languages, such as Claude Haiku 5.5 or Meta's Muse Ember (Notes: "The jobs agents repeat"); a person checking a sample; a GitHub repository with CSV files.
First step
Download the 2025 RASFF Window notices for pet food and the FDA's 2025 Veterinary enforcement entries, and map both to one set of columns.
Done looks like
At least 2,000 recalls and alerts from at least five regulators, 2010 to 2025, with a hand-checked sample of 200 rows whose fields match the source notice in at least 95% of cases, and a published count of hazards that led to recalls in more than one country within 12 months.
Open 02 · the brief
Read owner posts for outbreaks
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test, on past events, of whether public posts by pet owners would have flagged a pet food outbreak before the recall. For each event, such as the aflatoxin recall of December 2020 or the Hill's vitamin D recall of January 2019, collect the public posts that name the brand from the months before, group them by brand and symptom with a classifier, and record the date an alert would first have fired against the date of the recall. Publish counts only, never user names.
Why it is still open
Kaur and colleagues (Journal of Veterinary Internal Medicine, 2026) used AI models to sort about 24,000 owner posts from the first half of 2025, more than 96% from Reddit, on diet and health signs; they showed the method can capture owners' perceptions, not that it can catch outbreaks. The GitHub recall explorer found that when the first signs of a US problem were scattered complaints, the recall came 181 days to about 5.5 years later. No published test of public posts as an early warning for pet food outbreaks was found.
Built with what exists
Public posts gathered under a platform's research terms, such as Reddit's Data API, or a social listening service such as Pulsar, which Kaur's team used; a model that returns probabilities for fixed-answer questions about each post, such as Gemini Decision 1 (Notes: "A thousand questions, one pass"); a simple alert when posts about one brand rise above their usual level; Python.
First step
List ten pet food recalls since 2018 with their dates, and for one of them collect all public posts naming the brand from the three months before.
Done looks like
For at least five past events with known recall dates, the date an alert would have fired, the number of posts behind it and the false alarms over the same months, published with code. A useful result is an alert at least 14 days before the recall with fewer than one false alarm a month.