No. 010 September 27, 2026 9 min read

The Baseline Was Never a Floor

Atmospheric radiocarbon dropped below its pre-industrial level around 2020 and kept going, so a leaf grown in 2024 now reports roughly 175 radiocarbon years. The famous “by 2050, fresh material will look a thousand years old” is the worst of six scenarios — and a roadside leaf in Bangkok got there already.

The Australian and New Zealand practitioners’ guide to radiocarbon bomb-pulse dating opens its list of uses with something that sounds too simple to need a laboratory: bomb pulse dating “may be used to determine if the individual died before or after 1950.” A femur turns up in a drainage ditch. Crime scene or archaeological site? Burn a gram of collagen, count the carbon-14, and the answer comes back without anyone having to guess from the state of the bone.

That test has worked for about sixty-five years on an accident, and the accident is over. Not because the chemistry changed or anyone cut a corner. The atmosphere walked back across the line the whole scale was cut from, and kept walking.

The accidental instrument

Radiocarbon dating normally runs on decay: living things take up carbon-14 at the atmospheric ratio, stop when they die, and the clock starts. Resolution is set by the half-life, about 5,700 years, which makes the method excellent at centuries and hopeless at years. No amount of counting statistics turns a 5,700-year half-life into a one-year answer.

Atmospheric weapons testing did it instead. Above-ground detonations from the mid-1950s through 1963 flooded the troposphere with neutron-made carbon-14, and northern-hemisphere levels peaked in 1963–64, right as the Limited Test Ban Treaty came into force. Then the spike drained away — first into the surface ocean and the biosphere, which absorbed most of it within a few decades, and latterly for a different reason entirely.

The drain produced a steep, monotonic, well-surveyed curve, and a steep curve is a clock. Anything that grew on the descending limb can be placed on it. Tooth enamel, laid down once in childhood and never remodelled, gives year of birth to within one to two years; hair and nail give year of death to within one to three. A review of the bomb peak’s uses lists shark ages, human brain-cell birthdates, whisky authentication, and the result that mattered most in policy terms: roughly 90% of seized elephant tusks came from animals dead less than three years before confiscation — the difference between a stockpile being laundered and a poaching operation running right now.

−12.7 ‰

Δ¹⁴CO₂ at Jungfraujoch, 2024 annual mean

±2.0; it was −2.0 ‰ in 2019

−2.1 ‰/yr

the present rate of decline

±1.9; about 6 ‰/yr in 2000

~175

radiocarbon years a 2024 leaf now reports

my arithmetic, not a published figure

1.5 ‰

that lab's measurement precision on ¹⁴CO₂

the curve now moves 1.4× that per year

The line was never a floor

The unit everyone works in is F¹⁴C, the sample’s carbon-14 content as a fraction of a 1950 standard, and for sixty-five years its most useful property was the sign of one simple comparison. Above 1.0 meant bomb carbon, meant recent. Below meant old. That was the whole trick.

It stopped being true around 2020. A measurement programme at Jungfraujoch, the Swiss high-alpine station at 3,580 m, reports Δ¹⁴CO₂ annual means falling from −2.0 ± 3.8 ‰ in 2019 to −12.7 ± 2.0 ‰ in 2024, with every annual mean from 2020 onward below zero and a trend of −2.1 ± 1.9 ‰ per year. Zero, here, is roughly the pre-industrial atmosphere. The bomb signal did not merely fade back to the natural level; it went under it.

The reason it went under is the older story. Fossil carbon is millions of years old and contains no carbon-14 at all, so burning it dilutes the atmospheric ratio — the Suess effect, described in 1955, before the bomb peak had crested. Measurements in western European background air show the decline slowing from about 6 ‰ per year at the start of this century to about 3 ‰ per year by 2009–2012, and the authors are explicit about what the remaining slope is made of: the bomb excess “has been almost fully equilibrated with surface ocean water and the terrestrial biosphere.” What is left is combustion.

Run −12.7 ‰ through the conventional age formula and a leaf grown in 2024 reports something like 175 radiocarbon years. That is arithmetic on their number, not a figure anyone published, and the conversion is mechanical: F14C = (1 + D14C/1000) * exp(-(y-1950)/8267), then age = -8033 * ln(F14C). Doing the same for 2019 gives about 83 years. So the living atmosphere is currently gaining radiocarbon age at roughly eighteen years per calendar year, and about seventeen of those eighteen are dilution rather than anything decaying.

A fraction above one meant recent. That was the whole trick, and it expired around 2020.

What actually surprised me

I went in expecting the story to be the projection, and it isn’t. The projection is well known: a 2015 paper put it starkly — “by 2050, fresh organic material could have the same ¹⁴C/C ratio as samples from 1050, and thus be indistinguishable by radiocarbon dating” — and that sentence has been repeated ever since, usually without the conditional and always without the scenario.

The scenario matters a great deal. The same group’s 2020 update runs six CMIP6 pathways, and the spread is enormous: Δ¹⁴CO₂ hovers near 0 ‰ throughout under SSP1-1.9, bottoms out at −38 ‰ in the 2050s under SSP1-2.6, and reaches −105 ‰, −209 ‰ and −290 ‰ by 2100 under SSP2-4.5, SSP3-7.0 and SSP5-8.5 respectively. The 2050 date belongs to the top of that range. Take the middle and run the same conversion: −105 ‰ in 2100 works out to about 1,040 radiocarbon years, which puts the thousand-year milestone half a century later than the sentence everyone quotes.

Deepest projected Δ¹⁴CO₂ depletion this century, ‰ below the pre-industrial level

SSP1-2.6

−38 ‰ · 2050s

SSP2-4.5

−105 ‰ · 2100

SSP3-7.0

−209 ‰ · 2100

SSP5-8.5

−290 ‰ · 2100

But while reading around the projections I found something better, in a paper I nearly skipped because of its title. A students’ project — over 150 samples of leaves, flowers and grass collected across several continents for a decade, measured to about 0.003 in F¹⁴C — set out to watch the bomb peak fade in real material. Most of their samples sit between −0.7% and 4% fossil-fuel dilution relative to the alpine background. Some do not: traffic-adjacent sites run to 10%, Bangkok came in at 10.5%, Trieste at 5.6%.

A 10.5% dilution is roughly a 105 ‰ depression on top of the background. Put that through the same arithmetic and a leaf grown beside a road in Bangkok in the early 2020s carries something like 1,070 radiocarbon years — the same side of the line as the AD 1050 in that famous sentence. The 2050 headline is not a forecast there. It already happened, and it happened in a plant you could pick on your way to lunch.

That reframes the whole thing for me. The number everyone quotes is a global mean, and a global mean is the wrong object entirely when the question is whether one particular sample can be dated. What breaks bomb-pulse dating first is not the background trend but the variance around it — and the variance is already wider than fifty years of trend.

Where this gets weaker

Several places, and it is worth being plain about them.

The urban-leaf comparison is mine, and it does not transfer straight to bodies. Human tissue is built from a food supply grown elsewhere, averaged over a continent and a shopping season. Roadside air is a local extreme that human carbon mostly does not see.
The dilution percentages come with the authors’ own caveat: they applied no correction for other carbon dioxide sources such as decomposing organic matter, and they treat one alpine station as the global background for every sample.
The contamination runs both ways. The Jungfraujoch group had to correct for European nuclear power plants, which emit carbon-14 and push the signal up — at 3,580 m, on a mountain, in clean air.
I could not get past the paywall on the 2022 correspondence that first flagged the crossing, so my date for it comes from secondary coverage, which puts the dip below pre-bomb values in 2021. The alpine record’s own annual means suggest 2019–2020 and sit within their error bars either way. “The year it crossed” is not as sharp a fact as it sounds.
And the field’s own reviews are not tracking this well. A 2025 systematic review of radiocarbon for post-mortem interval estimation in bone gathered ten papers, found lag times between tissue and death ranging from 3.2 to 59.5 years, concluded the range “is not narrow enough to be useful for investigations” — and never discusses the closing window at all. The method’s best-documented weakness and its terminal problem are being written up separately.

Why I keep thinking about it

There is a partial rescue, and it is elegant. Fossil fuels are depleted in carbon-13 as well as carbon-14, so combustion leaves a second, independent fingerprint: measure both isotopes and a distorted δ¹³C tells you the ¹⁴C age has been pushed by fossil carbon. The process that breaks the clock signs its own work. It fails where carbon exchanges slowly — the deep Pacific is the stated example — and large-scale bioenergy with carbon capture would complicate it further.

What stays with me is the shape of the thing. For about sixty-five years, forensic science had a free, global, unrepeatable calibration signal, injected by the least scientific process imaginable, precise in a way the underlying physics flatly does not permit. Nobody built it, and nobody can renew it — you would have to resume atmospheric testing.

And it does not switch off on a date. Newly produced material loses the marker first, which is why the 2020 update says bluntly that “new ivory will soon also measure below 0 ‰, eliminating the use of ¹⁴C as a detection tool for illegal ivory.” Bone is different: someone who lived through the 1960s still carries bomb carbon in their skeleton, so the before-or-after-1950 test keeps working on them as long as their remains last. The capability is not being withdrawn so much as aging out with a cohort.

The part I find hardest to hold is that nothing went wrong. No instrument drifted, no paper needs retracting. A window opened by weapons testing is being closed by combustion, and the closing is done by the same accumulated emissions that everything else about this century turns on — so the clock is one more thing downstream of that, and a small one. It is only that it was, briefly, the most precise clock anyone had for recent time, and the contrast is going out of it in numbers you can read off a mountain, two per mil a year.