For most of the last century, "bird flu" was something that happened to birds, and occasionally to the people who worked closely with them. In 2026 the story is more complicated. The H5N1 virus has spread into dairy cattle across the United States, killed marine mammals by the thousands in South America, and infected dozens of farm workers — while, crucially, still failing to do the one thing that would turn it into a human pandemic: spread easily from person to person.
That gap between what the virus has done and what it could do is exactly why public-health agencies are watching so closely. This article lays out where things actually stand — separating confirmed facts from open questions, animal outbreaks from human risk, and "what's possible in theory" from "what's measured right now." The short version: for the general public, the current risk is low. The longer version is worth your time.
To keep those distinctions straight, the sections below move outward in rings — from the animals where the virus lives, to the handful of people it has reached, to the food on your table, to the machinery built to catch it early if it ever changes. Read in that order, the frightening historical numbers and the reassuring current risk assessment stop contradicting each other and start fitting together into one coherent picture.
Table of Contents
- Why H5N1 is back in the headlines
- Where the virus is now: birds, cattle, and other mammals
- Why the spillover into wild mammals matters
- Human cases: what the numbers say, and what they don't
- What the U.S. cases actually looked like
- What the global tally does and doesn't tell you
- Is the food supply safe? Milk, eggs, and pasteurization
- The milk on store shelves
- Where the real risk is: raw milk and undercooked food
- Preparing without panicking: vaccines, stockpiles, surveillance
- A closer look at each layer of defense
- So how worried should you be — and what to watch
- The specific signals scientists are watching for
Why H5N1 is back in the headlines
H5N1 is a subtype of highly pathogenic avian influenza (HPAI) — a flu virus adapted to birds that can cause severe disease in poultry. The specific version driving today's events belongs to a genetic branch called clade 2.3.4.4b, which since around 2021 has fueled the largest bird-flu outbreak ever recorded, a "panzootic" spanning Europe, Africa, Asia, and both American continents [source: US CDC, 2026].
The word "panzootic" is the animal-kingdom counterpart of "pandemic," and it is not an exaggeration here. Since roughly 2021 and 2022, clade 2.3.4.4b has pushed into Europe, Africa, Asia, North and South America, and even reached the Antarctic region — the widest geographic sweep ever documented for an avian influenza [source: US CDC, 2026]. Wild birds are the engine of that spread: they carry the virus between continents along their migratory routes, which is why no single country can simply fence the outbreak out at its border [source: US CDC, 2026].
Two developments pushed it from a veterinary concern into a global-health headline. First, in March 2024 the virus was confirmed for the first time in dairy cattle in the United States — a brand-new mammalian host, carrying high viral loads in udder tissue and raw milk [source: USDA APHIS, 2026]. Second, the count of human infections tied to those animal outbreaks kept climbing, and in January 2025 the United States recorded its first H5N1 death [source: US CDC, 2025].
The cattle jump deserves emphasis, because it opened a genuinely new chapter. The virus involved is a reassortant genotype labeled B3.13, and its arrival in cows marked the first time H5N1 established cow-to-cow spread in a mammal rather than simply spilling over from birds into a dead-end host [source: USDA APHIS, 2026]. The high viral loads found in udder tissue and raw milk are what make dairy operations — and the milking process in particular — a plausible route for the virus to move from one animal to the next.
None of this means a pandemic is underway. It means the virus is doing more, in more species, in more places, than it used to — and each new mammalian host is a fresh opportunity for it to adapt. Understanding the difference between "wider spread in animals" and "a change in human risk" is the single most important idea in this whole story.
Where the virus is now: birds, cattle, and other mammals
Start with the animals, because that is where nearly all the action is.
- Wild birds and poultry. Clade 2.3.4.4b is now entrenched in wild bird populations, which carry it across continents along migratory routes. In the United States, more than 168 million poultry have been affected or culled across all 50 states since early 2022, because standard containment means depopulating an entire infected flock [source: USDA APHIS, 2026].
- Dairy cattle. Since the first Texas detection in March 2024, the virus — a reassortant genotype labeled B3.13 — has been confirmed in more than 1,100 herds across at least 19 U.S. states as of mid-2026 [source: USDA APHIS, 2026]. Some dairy states, such as Idaho, have been hit especially hard. Cow-to-cow spread is thought to occur largely through the milking process and animal movement, though not every transmission route has been fully explained.
- Other mammals. The virus has been detected in seals, sea lions, whales, foxes, bears, skunks, domestic cats, and mink. In South America it caused mass die-offs of sea lions and seals, and some infected mammals showed neurological damage [source: WOAH, 2026].
That poultry figure is worth pausing on, because it explains why the outbreak keeps making the news even in years when human cases are rare. The number is so large precisely because containment is blunt: when infection is confirmed on a commercial farm, the standard response is to depopulate the entire flock, so a relatively small number of detections can translate into millions of culled birds [source: USDA APHIS, 2026]. The cattle picture is different in kind, not just in scale — here the concern is not only the loss of animals but the fact that the virus appears able to move from cow to cow, most likely through shared milking equipment and the movement of animals between farms, even though investigators have not mapped every path [source: USDA APHIS, 2026].
Two clarifications matter here. Different genotypes are circulating in parallel: the D1.1 lineage moving through wild birds and poultry is distinct from the B3.13 lineage in cattle. And "found in mammals" is a statement about animal epidemiology, not about human transmission — the leap from an infected cat to a sustained human outbreak is enormous, and has not happened.
Why the spillover into wild mammals matters
The list of infected mammals is not just a curiosity. In South America, clade 2.3.4.4b tore through marine-mammal colonies, causing mass die-offs of sea lions and seals in countries including Peru, Argentina, and Chile, and some infected animals showed signs of neurological damage [source: WOAH, 2026]. Every one of those species is a mammal, and each mammal the virus infects is an environment slightly more like our own than a bird is. That does not mean any of these animals are handing the virus to people — in any sustained way, they are not — but it is the reason scientists treat the widening mammalian host range as a signal worth tracking rather than a footnote. The concern is about opportunity accumulating over time, not about a route to humans that exists today.
Human cases: what the numbers say, and what they don't
This is where careful reading matters most, because the raw figures can be pulled in two opposite and equally misleading directions.
What the U.S. cases actually looked like
In the United States, roughly 71 human H5 infections were confirmed from 2024 through 2026, almost all in dairy or poultry workers with direct exposure to infected animals. No sustained person-to-person transmission has been identified [source: US CDC, 2026]. Many of these infections were caught precisely because health officials were actively monitoring exposed farm workers — a sign the surveillance net is working, not that the virus is spreading unseen. That near-universal link to direct animal contact is itself a reassuring signal: if the virus were quietly moving between people, infections would be surfacing in the wider community rather than clustering among the exposed [source: US CDC, 2026]. Most of these U.S. cases were mild — often conjunctivitis (pink eye) or mild respiratory symptoms.
But mild is not the whole picture. In January 2025 the CDC reported the first U.S. death: a person over 65 with underlying medical conditions, infected after exposure to a backyard flock and wild birds, and carrying the D1.1 genotype rather than the cattle strain [source: US CDC, 2025]. That fatal case was in Louisiana, and it was announced on January 6, 2025 [source: US CDC, 2025]. In November 2024, California had confirmed the first U.S. pediatric case, who recovered from mild illness; no community spread was found [source: US CDC, 2026]. For that child, the exact source of exposure was not clear at first, even though investigators never found evidence of spread within the community [source: US CDC, 2026].
Notice how much the genotype and the circumstances vary across these cases. The mild dairy-worker infections have generally involved the B3.13 cattle strain and close animal contact, while the fatal Louisiana case involved the D1.1 genotype, older age, and underlying illness [source: US CDC, 2025]. Severity, in other words, tracks with a mix of the viral genotype, the route and intensity of exposure, and the patient's own health — which is exactly why a single death cannot be read as a population-wide fatality rate, and why a run of mild cases cannot be read as proof the virus is harmless [source: US CDC, 2026].
What the global tally does and doesn't tell you
Globally and historically, the numbers look frightening in isolation: from 2003 to early 2026, WHO recorded 993 confirmed human H5N1 cases across 25 countries, with 477 deaths — a case-fatality ratio near 48% [source: WHO, 2026]. That figure deserves a giant asterisk. It counts mostly people sick enough to seek hospital care, so milder and asymptomatic infections are almost certainly undercounted, which inflates the apparent lethality. A high historical case-fatality ratio tells you H5N1 can be severe; it does not tell you your personal odds today. The contrast with the recent U.S. experience is instructive: the American cases were caught mainly through active monitoring of exposed workers rather than hospital admissions, and they skewed mild — which is exactly what you would expect when you go looking for infections instead of waiting for the sickest patients to arrive [source: US CDC, 2026].
The most recent human cases outside the United States fit the same pattern of animal-linked, non-spreading infection. In early 2026 the WHO tracked additional H5N1 cases, including a fatal pediatric case in Bangladesh and several cases in Cambodia [source: WHO, 2026]. The Cambodian cases are worth distinguishing carefully: they stem from a different genetic lineage — clade 2.3.2.1c, a reassortant that has continued to cause human infections there — rather than the 2.3.4.4b clade behind the American cattle outbreak [source: WHO, 2026]. In every one of these instances, the infections were linked to animal or environmental exposure, with no sustained human-to-human chains [source: WHO, 2026].
The honest synthesis avoids both traps. "Most cases are mild" does not mean the virus is harmless; "the historical fatality rate is ~48%" does not mean half of infected people now will die. The load-bearing fact sits between them: the virus can cause severe disease, but it currently lacks the ability to spread efficiently between humans.
Is the food supply safe? Milk, eggs, and pasteurization
Because the virus turned up in dairy cattle, the obvious question is whether it is in the food you buy. Here the evidence is reassuring and specific.
The milk on store shelves
The U.S. FDA ran multiple surveys of retail dairy products. Its first survey found no live, infectious H5N1 in any of 297 samples; a second survey of 167 products collected between June 18 and July 31, 2024 — pasteurized milk, cheese, butter, ice cream, and aged raw-milk cheese — also found no viable virus [source: US FDA, 2025].
The crucial nuance is the difference between detecting genetic material and detecting a living virus. Early retail-milk testing did find fragments of viral RNA in a substantial share of samples — around one in five by FDA's count, and about 36% in a later peer-reviewed study [source: CDC EID Journal, 2026]. But when scientists tried to grow live virus from those samples, or used them to infect eggs and mice, nothing infectious emerged. A positive PCR result flags the trace of a virus, not a virus capable of infecting you — and it is evidence that pasteurization did its job of inactivating the pathogen. The FDA separately ran a laboratory study using commercial pasteurization equipment, which likewise confirmed that the process reliably inactivates the virus [source: US FDA, 2025].
It helps to understand why those two kinds of tests give different answers. A PCR test amplifies tiny snippets of the virus's genetic code, so it will light up even if all it finds are broken, non-infectious fragments — the molecular equivalent of finding a fingerprint rather than the person who left it. Trying to grow the virus in cell culture, or to infect eggs and mice with it, is the far more demanding test of whether anything alive and capable of replicating remains; on the pasteurized retail samples, that test kept coming back negative [source: CDC EID Journal, 2026].
Where the real risk is: raw milk and undercooked food
The exception is raw, unpasteurized milk, which can contain infectious H5N1; the FDA and CDC advise against consuming it [source: US FDA, 2026]. This is not a purely theoretical caution: in animal experiments, animals fed raw milk from infected cows became infected themselves, which is the practical reason the raw product is treated so differently from the pasteurized one [source: US FDA, 2026]. Properly cooked poultry and eggs are also considered safe, and infected flocks are culled before entering the food supply [source: US CDC, 2026]. "Properly cooked" has a specific meaning here — an internal temperature of 165°F (74°C) — and at that point the ordinary rules of food safety do the work [source: US CDC, 2026]. In short: pasteurization and cooking are the safeguards, and they are working.
Preparing without panicking: vaccines, stockpiles, surveillance
Preparedness is often misread as a signal that danger is imminent. It is the opposite: it is what lets societies keep the current risk low and respond fast if that changes.
In May 2026, the FAO, WHO, and WOAH jointly assessed the global public-health risk from H5N1 to the general population as low, and the risk to occupationally exposed people — dairy and poultry workers — as low to moderate, depending on the protective measures in place [source: FAO/WHO/WOAH, 2026]. Animal-to-animal spread continues, human infections remain limited, and sustained human-to-human transmission has not occurred. These agencies also revisit that judgment on a regular schedule rather than issuing it once, so any real shift in the animal or human data would surface in an updated assessment.
That "low to moderate" band for workers is doing quiet but important work. It says the everyday risk to the public and the occupational risk to someone milking infected cows are not the same number, and that the worker's end of the range can be pulled down by the right protective measures rather than being fixed by fate [source: FAO/WHO/WOAH, 2026].
Behind that assessment sits real infrastructure:
- Candidate vaccine viruses. The WHO maintains a library of candidate vaccine viruses (CVVs) for zoonotic flu, including matched A(H5N1) strains, with the most recent A(H5) update published in February 2026. These are re-evaluated at least twice a year and can serve as a head start if a vaccine is ever needed at scale [source: WHO, 2026].
- National stockpiles. In the U.S., the agencies ASPR and BARDA maintain a pre-pandemic influenza vaccine stockpile with H5 antigen matched to clade 2.3.4.4b, alongside manufacturing contracts. This is contingency capacity, not a rollout — there is no general H5N1 vaccination campaign underway [source: US ASPR/BARDA, 2026].
- Surveillance. The CDC monitors exposed workers, watches influenza A signals in wastewater, and leans on the seasonal-flu surveillance network to catch anything unusual early — such as a mutation improving the virus's ability to bind human cells [source: US CDC, 2026].
A closer look at each layer of defense
Each of those layers is more deliberate than it looks. The candidate vaccine viruses are not a finished product sitting on a shelf; they are reference strains, kept genetically matched to the circulating virus, that a manufacturer could use as a starting point. The WHO refreshes them on the same rhythm it uses for seasonal flu — the February 2026 A(H5) update was issued alongside the consultation that sets the composition of the 2026–2027 Northern Hemisphere flu vaccines — and re-evaluates them at least twice a year, and whenever the virus shifts [source: WHO, 2026].
The stockpile is similarly a hedge rather than a plan already in motion. The U.S. pre-pandemic reserve holds H5 antigen matched to clade 2.3.4.4b, and the government maintains supply and fill-finish contracts with established influenza-vaccine makers — CSL Seqirus, Sanofi, and GSK — including adjuvanted H5 formulations [source: US ASPR/BARDA, 2026]. All of it is contingency capacity; there is no general public H5N1 vaccination campaign underway [source: US ASPR/BARDA, 2026].
Surveillance is the layer most likely to give the first warning. Beyond monitoring exposed workers, the CDC watches for influenza A signals in wastewater and leans on the existing seasonal-flu network to flag anomalies, with particular attention to the specific genetic changes that would matter most: mutations that adapt the virus to mammals, that improve its ability to bind human cell receptors, or that confer resistance to antiviral drugs [source: US CDC, 2026].
Preparedness, in other words, is the quiet machinery that turns "a low but non-zero risk" into "a risk we can act on."
So how worried should you be — and what to watch
Put the pieces together and the current picture is coherent. For the general public, the measured risk right now is low: there is no sustained human-to-human transmission, nearly all human cases trace to direct animal contact, and the everyday exposure routes people worry about — pasteurized milk, cooked food — have been tested and cleared [source: FAO/WHO/WOAH, 2026].
So why keep watching at all? Because influenza is a shapeshifter. Flu viruses evolve both by gradual mutation and by "reassortment" — swapping whole gene segments when two flu viruses infect the same host. Every additional mammal H5N1 infects, from dairy cows to sea lions, is another roll of the dice for the virus to acquire human-adapting changes. That is not a prediction that a pandemic will happen; it is the reason surveillance exists — to lower the odds and to buy time. History earns the caution: the 1918, 1957, 1968, and 2009 flu pandemics all emerged from animal-origin viruses that reassorted [source: WHO, 2024].
That history is worth naming precisely, because it shows a pattern rather than a single scary precedent. The 1918 pandemic was an H1N1 virus, 1957 was H2N2, 1968 was H3N2, and 2009 was a new H1N1 — four different subtypes, each arising when animal-origin influenza reassorted into something humans had little immunity to [source: WHO, 2024]. H5N1 is watched so closely precisely because it is the kind of virus that could, in theory, one day follow that template — while, just as importantly, it has not done so [source: WHO, 2024].
The specific signals scientists are watching for
If the risk picture ever changes, it will change through a small number of concrete, detectable signals rather than all at once. The three that matter most are the appearance of efficient, sustained human-to-human transmission; mutations that make the virus better at infecting people — for example, changes that improve its ability to bind human cell receptors or that adapt it to mammalian hosts; and a cluster of human cases with no animal link, which would be the clearest hint that the virus had begun spreading on its own [source: US CDC, 2026]. These are not hypothetical worries invented for this article; they are the exact things the CDC's monitoring of workers, wastewater, and the seasonal-flu network is designed to catch early [source: US CDC, 2026]. As of now, none of them has appeared, and nearly every confirmed human infection still traces back to direct contact with animals [source: US CDC, 2026].
For an ordinary reader, the practical takeaways are modest. Avoid raw milk and don't handle sick or dead birds and animals with bare hands. If you work with poultry or cattle, follow protective guidance, because your risk sits a notch higher than the public's. That added risk is manageable with basic precautions — gloves, eye protection, and respiratory protection when handling potentially infected animals [source: US CDC, 2026]. Beyond that, the things that would actually change the risk picture are the ones scientists are already watching for: evidence of efficient human-to-human transmission, mutations that make the virus better at infecting people, or a cluster of cases with no animal link. As of now, none of those has appeared — and that "as of now," reported plainly and updated often, is the most honest headline bird flu can carry in 2026.