Industries / Food, farming and forestry / Food and beverage processing
Food safety and testing
Food safety and testing is the set of rules, plant routines and laboratory checks that keep processed food free of germs, stray objects, hidden allergens and fakes, and that trace an illness back to the food that caused it.
7
Milestones achieved, from 1963 to 2019.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1963
A shared world food code
The UN's Food and Agriculture Organization (FAO) and World Health Organization (WHO) set up the Codex Alimentarius Commission, a body of governments that writes food standards any country can adopt. It first met in 1963 and now has 189 members: 188 countries and the European Union.
Credit
FAO and WHO
1971
HACCP shown to industry
HACCP (Hazard Analysis and Critical Control Points) is a method in which a plant lists what could go wrong at each step, then sets a measurable check at the few points that control each hazard. Pillsbury developed it in the 1960s with NASA to make space food safe, building on methods from the US Army's Natick laboratories. A 2018 review says it was presented to the food industry for the first time at the first National Conference on Food Protection in spring 1971. Codex adopted HACCP guidelines in 1993, and the US required HACCP in meat and poultry plants from 1996.
Credit
Pillsbury, in a team led by Howard Bauman; NASA; US Army Natick laboratories
1994
A germ ruled an adulterant
In 1992 and 1993, undercooked hamburgers from the Jack in the Box restaurant chain caused an outbreak of E. coli O157:H7. The CDC's report at the time counted more than 500 laboratory-confirmed infections and four deaths. In 1994 the US Department of Agriculture's Food Safety and Inspection Service (FSIS) declared raw ground beef carrying the germ "adulterated", which means it cannot legally be sold.
Lessononce a germ in raw food is illegal, keeping it out becomes the plant's job, not only the cook's.
Credit
USDA FSIS, under administrator Michael Taylor
1996
Labs share DNA fingerprints
The CDC started PulseNet, a network of laboratories that compare the DNA "fingerprints" of bacteria from sick people, so that cases in different states can be linked to one food. The CDC says it grew out of the 1993 E. coli outbreak. It now has 83 public health and food regulatory laboratories, at least one in every state.
Lessonan outbreak spread thinly across many places stays invisible until laboratory results are pooled.
Credit
CDC and state public health laboratories
2008
Melamine fooled the protein test
In China, melamine, an industrial chemical rich in nitrogen, was added to milk to raise the apparent protein content in tests that measure nitrogen. China reported 294,000 children affected and 6 deaths. Sanlu Group infant formula held up to 2,563 mg of melamine per kg.
Lessona test that measures a stand-in, nitrogen, instead of the real thing, protein, invites cheats who fake the stand-in.
Credit
figures reported by China, summarized in Environmental Health Perspectives (2009)
2013
DNA tests exposed horsemeat
In January 2013 the Food Safety Authority of Ireland found horse DNA in 10 of 27 beef burger products. UK tests then found horse meat in burgers, lasagne and bolognese ready meals, and the UK Food Standards Agency required food businesses to run authenticity tests on processed beef. A government review by Chris Elliott led to the National Food Crime Unit, set up in May 2015.
Lessonchecking that the right ingredient is present does not show that the wrong one is absent, and DNA tests check the species directly.
Credit
Food Safety Authority of Ireland; UK Food Standards Agency; Chris Elliott of Queen's University Belfast
2019
Genome sequencing replaces fingerprints
PulseNet laboratories began moving from the older fingerprint method, which cuts DNA into a pattern of bands, to whole genome sequencing, which reads a germ's entire genetic code. The FDA had started GenomeTrakr, a network of laboratories sequencing germs from food and the environment, as a pilot in 2012. In a 2018 Salmonella outbreak linked to ground beef, sequencing showed that not all cases that looked alike were linked to the beef.
Lessona finer test separates cases that looked alike.
Credit
CDC PulseNet; FDA GenomeTrakr
Watch for
Listeria swabs in every district
FSIS says its inspectors will add four surface swabs to routine sampling of ready-to-eat meat, two from surfaces that touch food and two from other surfaces, tested for Listeria. A pilot ran in fiscal year 2026 in its Alameda, Chicago and Raleigh districts, and the plan is to phase it in across all nine districts from January to March 2027. FSIS cites its investigations of the 2024 listeriosis outbreaks.
Achieved whenall nine districts are swabbing and FSIS publishes the results.
Claimed by
USDA FSIS, September 2026
Food traceability rule postponed
The FDA's Food Traceability Rule of 2022 requires firms handling foods on a high-risk list to keep set records at each step, from harvesting and cooling to packing, shipping, receiving and processing. In August 2025 the FDA proposed moving the compliance date by 30 months, from 20 January 2026 to 20 July 2028, and Congress told it not to enforce the rule before then.
Achieved whenenforcement begins and the FDA reports covered firms producing these records in a real outbreak investigation.
Claimed by
FDA, August 2025
Cameras spot plastic in mince
INNDEO, a Spanish inspection-machine maker using cameras from Specim, says its system detects "most plastic foreign bodies on the product surface" of minced meat at conveyor speed. It uses hyperspectral cameras, which record many narrow bands of light so that materials can be told apart by their light signature. X-ray machines and metal detectors depend on density or conductivity, so low-density plastic is nearly invisible to them. The case study gives no detection rates and names no plant.
Achieved whenan independent laboratory or a plant publishes detection and false-reject rates for stated plastic types and sizes.
Claimed by
Specim with INNDEO, August 2026
Open
01
Robot that takes swabs
A mobile robot arm that moves along a mock processing line and swabs 50 numbered sites, such as belts, frame legs, drain grates and the undersides of guards. It wipes each site at a set pressure and pattern, seals and labels the swab, and logs site, time and force. The sites are soiled only with harmless material, such as diluted meat juice, milk or a fluorescent tracer, and read with ATP swabs. ATP is a molecule found in all living cells and food residue, and an ATP swab with a hand-held meter gives a reading in seconds. No germs that cause disease are used.
The briefCredit · unclaimed02
Swab plan from a walk-through
A tool that takes a phone video of a processing room and its equipment list and drafts a monitoring plan. The plan is a numbered list of swab sites with photos, ordered from food-contact surfaces outward. It flags likely hiding places for germs, such as hollow rollers, cracked floors, condensation drip points and hard-to-reach joints, and suggests how often to swab each one.
The briefCredit · unclaimed
Open 01 · the brief
Robot that takes swabs
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A mobile robot arm that moves along a mock processing line and swabs 50 numbered sites, such as belts, frame legs, drain grates and the undersides of guards. It wipes each site at a set pressure and pattern, seals and labels the swab, and logs site, time and force. The sites are soiled only with harmless material, such as diluted meat juice, milk or a fluorescent tracer, and read with ATP swabs. ATP is a molecule found in all living cells and food residue, and an ATP swab with a hand-held meter gives a reading in seconds. No germs that cause disease are used.
Why it is still open
Plant swabs are still taken by people, and FSIS's 2027 expansion adds swabbing tasks for its inspectors. The closest work is a University of Arkansas study from September 2025, in which a Universal Robots UR5e arm swabbed a test pad and picked up 98.4% of the test liquid, against 94.8% for trained people, with steadier pressure. The authors say a version on a mobile robot is in progress and that validation in real production is future work. No shipped swabbing robot turned up in searches for this file.
Built with what exists
A cobot such as the UR5e on a mobile base that maps a room and drives to saved spots (the note "A camera on patrol" covers a consumer robot that does this), or a low-cost arm run with Hugging Face's LeRobot. A robot model that learns from a few demonstrations, such as NVIDIA's GR00T 2 (see the note "Five demonstrations per task"). Force sensors in the swab holder, commercial sponge swabs and a commercial ATP meter.
First step
Build a 10-site rig from stainless-steel offcuts, soil it with diluted meat juice, and record a technician's ATP readings and five demonstration swabs per site.
Done looks like
On three separate days the robot swabs all 50 sites with no one stepping in. Each site has a twin soiled the same way, which a trained technician swabs. The robot's ATP pickup is at least as high as the technician's at 45 or more of the 50 site pairs, and its day-to-day spread per site is smaller. The data and video are published.
Open 02 · the brief
Swab plan from a walk-through
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A tool that takes a phone video of a processing room and its equipment list and drafts a monitoring plan. The plan is a numbered list of swab sites with photos, ordered from food-contact surfaces outward. It flags likely hiding places for germs, such as hollow rollers, cracked floors, condensation drip points and hard-to-reach joints, and suggests how often to swab each one.
Why it is still open
A 2019 Cornell release says plant staff "rely on their own expertise and knowledge of the building to determine where to swab". Cornell's own model of where Listeria is likely to be needs detailed inputs on equipment, cleaning schedules and traffic, not pictures. A 2024 paper from the International Dairy Foods Association advises bringing in an expert to decide what to test, where and how often. No tool that drafts a plan from images turned up in searches for this file.
Built with what exists
An open-weights vision-language model, such as Qwen2.5-VL-7B or a small model fine-tuned on one GPU, run on a local workstation so that plant video stays on site, a phone camera, and a spreadsheet as the output.
First step
Ask a university food science pilot plant for permission to film one room and to have two staff mark their swab sites, then run an off-the-shelf vision-language model on the footage as a baseline.
Done looks like
For five rooms, for example in university pilot plants, two experienced food safety professionals mark swab sites on floor plans independently. The tool's draft includes at least 80% of the food-contact sites that both experts chose, and the experts rate at least half of its other suggestions as worth swabbing. The marks and results are published.