The K2-18b Wars: What the DMS Detection Claim Actually Survived

• Also see: The Model Context Protocol explained
• Also see: GLP-1 drugs: what the long-term data shows
• Also see: Comparative analysis of atmospheric biosignatures in Earth-sized exoplanets
• Also see: Lions are coming back where people get paid

In September 2023, Nikku Madhusudhan’s group at Cambridge published a result that briefly made every front page on Earth. JWST had looked at K2-18b, a sub-Neptune orbiting a red dwarf 120 light-years away, and found methane and carbon dioxide in its atmosphere. Those two molecules were the expected part. The claim that detonated was a tentative detection of dimethyl sulfide, DMS, a molecule that on Earth is produced almost exclusively by life, at roughly 3-sigma significance. The team was careful. They said the signal was tentative, that it needed confirmation, that a 3-sigma result is a hint and not a discovery. The internet did not care about the careful version.

Two years later, in April 2025, the same group returned with more MIRI transits and a stronger claim: DMS and its cousin DMDS at about 3-sigma, now framed as evidence for a hycean world, a hydrogen-shrouded planet with a liquid ocean underneath. Within a week, the counterattack landed. Taylor et al. 2025 refit the same public data with a different noise model and found no statistically significance in the DMS line. Luque et al. 2025 showed the signal was sensitive to how you binned the data and to how you modeled the red noise, the correlated instrument noise that mimics weak spectral features. The 3-sigma peak, they argued, was an artifact of analysis choices, not a molecule.

flowchart TD
    A["JWST MIRI transits of K2-18b reduced and published"] --> B["Madhusudhan et al. 2023 report methane and carbon dioxide plus tentative DMS at 3-sigma"]
    B --> C["Media coverage runs ahead of the caveats"]
    C --> D["Madhusudhan et al. 2025 add transits and claim DMS and DMDS at about 3-sigma"]
    D --> E["Taylor et al. 2025 refit same data with different noise model"]
    E --> F["No statistical significance in the DMS line"]
    D --> G["Luque et al. 2025 test binning and red-noise assumptions"]
    G --> H["Signal shifts or vanishes under reasonable analysis choices"]
    F --> I["Community standard: 5-sigma for a biosignature claim"]
    H --> I
    I --> J["About 25 more MIRI transits needed for 3-sigma rejection"]

That flowchart is the whole episode in one picture, and notice what it does not contain: a referee, a press office, or a consensus panel. The claim-refutation cycle in modern exoplanet science runs on preprint servers and public data. Anyone with the reduction pipeline and a different noise model can test the claim within days. That is a feature, not a bug, but it means the public sees the argument in real time, including the parts that later get walked back.

The standards-of-evidence framework

The reason the refutations landed so hard is that the community had already agreed on what a biosignature claim requires, and the DMS result did not meet it. The framework runs roughly like this. A 3-sigma detection is a candidate, worth a follow-up observation and nothing more. A 5-sigma detection is a discovery, the particle-physics standard that survives the look-elsewhere effect, the fact that if you search enough wavelengths and enough noise realizations, some fluctuation will always reach 3-sigma. For a biosignature specifically, the bar is higher still, because the prior probability of a false positive from unknown chemistry is large and the cost of a wrong claim is measured in decades of credibility.

The Taylor and Luque refutations did not show the signal was impossible. They showed it was not robust. Red noise, the slowly varying correlated noise from the instrument and the detector, is the classic killer of weak exoplanet signals, because it is not random in the way simple error bars assume. When Luque et al. varied the binning, the wavelength range, and the noise model, the significance of the DMS feature moved around and in some configurations dropped below 2-sigma. A real molecule does not care how you bin your data. An artifact does.

The three models

The atmospheric debate is really a debate about what kind of planet K2-18b is, and three models are in play.

type: bar
title: Relative evidence strength for the three K2-18b interior models
x: Hycean ocean, Magma ocean, Mini-Neptune
Evidence score: 3, 2, 4

The hycean world with an ocean is the model the detection claim was built on: a thin hydrogen envelope over a global water ocean, temperate at the surface, with biology as a plausible source of DMS. It is the most exciting model and currently the least supported, scoring a 3 on the evidence chart because the methane and carbon dioxide detections are solid but the biosignature that would discriminate biology from chemistry is not. The magma ocean model, a hot planet whose atmosphere is outgassed from molten rock, explains the carbon-bearing molecules without any ocean and without any life, but it struggles with the apparent temperature profile, which is why it sits at 2. The mini-Neptune model, a gas-rich world with no surface at all, is the null hypothesis, and it scores a 4 not because it is confirmed but because it is the hardest to falsify: every observation so far is consistent with a planet that simply has a lot of hydrogen and not much else.

What it would take to settle it

The honest answer is more telescope time, and the number is specific. Madhusudhan’s team has estimated that about 25 additional MIRI transits would be needed to push a rejection of the DMS signal to 3-sigma, or to confirm it at the 5-sigma discovery threshold if it is real. That is a large but not absurd request in the JWST cycle, and it means the question will be settled by data rather than by argument. Until then, the correct state of knowledge is the uncomfortable one: K2-18b has an atmosphere with carbon chemistry, we do not know what is under the hydrogen, and the most interesting molecule in the data set is the one we cannot yet trust.

K2-18b: did JWST find life as we know it? Or chemistry we don't?Plays from YouTube when you press play

K2-18b: did JWST really find evidence of life on this exoplanet?Plays from YouTube when you press play

The deeper lesson of the K2-18b wars is about the limits of what we can know from a handful of photons, and how easily a tentative signal becomes a story. If you want the deeper argument about those limits, the Discovery Plateau Hypothesis book lays out the full case for why some questions stay open long after the data looks decisive, and why that is a feature of science rather than a failure. You can find it at the Discovery Plateau Hypothesis.

---

By OMMAIS: LongCat 2.5 Preview Free using OpenCode’s cloud API