A steel that grows a second suit of armour out of manganese — the one element every textbook says wrecks corrosion resistance. It's real, it's peer-reviewed, and it's been announced as breaking news five times since 2023. The "40x cheaper" everyone quotes was never about hydrogen.

🔬 The element that is not supposed to work

Every metallurgist learns the same thing about manganese and stainless steel: it makes corrosion worse.

It’s in the textbooks. It’s in the alloy specs. Manganese is cheap, it helps you use less nickel, and it costs you corrosion resistance — that’s the trade, and it has been the trade for the better part of a century.

So when a PhD student in Hong Kong watched a manganese-bearing steel survive voltages that dissolve the best stainless alloys on the market, the first reaction was not excitement.

That is a lovely thing for a scientist to say out loud, and the result behind it is real. It’s peer-reviewed, it’s in a good journal, and the mechanism is genuinely new.

It is also, as of this week, three years old — and it has been announced as breaking news five separate times.

This issue is about both halves of that. The science deserves the attention. The way the attention arrived tells you something uncomfortable about how technology news reaches you in 2026.

🛡️ What the steel actually does

Stainless steel doesn’t resist corrosion because it’s dense or hard. It resists corrosion because it grows its own skin.

Chromium in the alloy reacts with oxygen and forms a film of chromium oxide — Cr₂O₃ — a few nanometres thick. Scratch it and it reforms in seconds. That film, and only that film, is why your kitchen sink isn’t a pile of rust.

The film has a ceiling, and the ceiling is electrical.

1. Push the metal to a higher and higher voltage and, at around 1000 mV (measured against a saturated calomel electrode), the stable Cr₂O₃ oxidises further — into soluble Cr(VI) species that simply wash away.

2. The skin dissolves. Metallurgists call this transpassive corrosion. Below that voltage the steel is immortal; above it, the steel is a sacrifice.

3. Splitting water needs about 1600 mV.

4. 1000 is less than 1600. That gap is why you cannot build the business end of a water electrolyser out of stainless steel.

5. So industry uses titanium coated in platinum or gold instead — which works beautifully and costs what you’d expect.

Even 254SMO, the benchmark super-austenitic alloy that shrugs off seawater pitting, hits the same wall. It’s a chromium problem, and every chromium-based alloy inherits it.

Here is what the Hong Kong team found. In their steel, a second film forms on top of the first one — a manganese-based layer that appears at around 720 mV and takes over exactly as the chromium layer approaches its limit. They call it sequential dual-passivation: two shields, deployed in order, each covering the other’s weakness.

The team spent close to six years on it — from the first anomalous reading to understanding it at the atomic scale. It’s the same Hong Kong group behind an antimicrobial stainless steel in 2021 and two ultra-strong “Super Steels” in 2017 and 2020. They are not amateurs and this is not a fluke.

That’s the good news, and it is properly good. A new passivation mechanism is a rare thing. If manganese can be made to do this reliably, it opens a design space that corrosion science had written off.

Now the part that made me stop.

📅 Then I checked the date

I found this story the way you probably did: a headline in a feed, sometime in the last two weeks. “Cannot be explained” — New super steel stuns scientists. Published 11 August 2026.

Read that article and you will find no date on the discovery. The verbs are present-perfect throughout: a team has developed; the findings were published in Materials Today; patents have been applied for, and two have already been authorised.

The date is in the article. It’s in the journal citation at the bottom, in small type, and it reads: Materials Today, 2023; 70: 8.

The paper went online on 18 August 2023. The university announced it on 14 November 2023. Everything you just read about manganese was known, published and press-released before the end of 2023.

Since then, the same press release has been rewritten as news in at least five distinct waves:

• Wave 1 — Nov–Dec 2023. The original, legitimate cycle: university release, ScienceDaily, trade press.

• Waves 2–4 — 2024 into spring 2025. Aggregator pickup, then a trade-press re-run, then another round.

• Wave 5 — May and August 2026. Two fresh ScienceDaily write-ups, each with a wide secondary pickup behind it.

Roughly half of all the coverage this 2023 paper has ever received was published in 2026.

And here’s the detail that I think is the actual story, because it shows someone knew:

I don’t know why. It may be nothing more than a shorter house format on a busy day. But the effect is not nothing: the August version is what the second wave of outlets picked up, and they picked it up clean of any date.

By the time it reached the far end of the chain it had also gained things the paper never said. At least one outlet ran “made from seawater.” Several ran hydrogen produced “directly from the ocean.” We’ll come back to that, because it matters more than it sounds.

🧮 The 40× is real. It is also not about hydrogen.

Every version of this story carries the same number: SS-H2 could cut costs by about 40 times.

Here is exactly what that number is, in the university’s own words:

Read the last four words again. The cost of structural material. Not the cost of the electrolyser. Not the cost of hydrogen.

Three boundaries get crossed every time this number is repeated:

• Structural material → the finished part. Machining, sintering, coating, forming and QA do not get 40× cheaper. Materials are only around 60% of stack cost.

• The stack → the installed plant. Engineering, procurement, construction and installation are typically half or more of installed capex. Structural parts are a large slice of the stack and a small slice of the plant.

• Capital cost → the price of hydrogen. Electricity is the dominant input. Capex is a minority share of the levelised cost.

Do the arithmetic all the way through and you get a hard ceiling. Even granting every assumption in the press release, and even if the structural material became free, the effect on the levelised cost of hydrogen lands in the single digits of percent — call it 5–9%, depending on the electricity price and capacity factor you assume.

That is a genuinely useful improvement! A 5% cut in green hydrogen costs is worth chasing. It is not the same sentence as “40 times cheaper,” and the second sentence is the one that travels.

One more thing worth flagging honestly. HK$17.8 million for 10 MW works out to about US$228 per kilowatt. For comparison, the IEA’s most recent hydrogen review puts PEM systems made and installed in China at US$600–1,200/kW, and outside China at US$2,000–2,600/kW. The US Department of Energy’s baseline for installed PEM capex is around $2,000/kW.

$228/kW isn’t in that range. It’s most likely a stack or equipment price rather than an installed-plant price — the phrase “electrolysis tank system” is a fairly direct translation of the Chinese term for the stack vessel, not the plant. That reading makes the number plausible. But the press release doesn’t say which, and it cites no source for either the HK$17.8m or the 53%.

That’s not misconduct. It’s a press office writing a cost claim into a corrosion-science story. The paper is about passivation mechanisms; it isn’t a techno-economic analysis, and it doesn’t pretend to be. The estimate got attached at the communications stage and has been repeated verbatim, unchecked, for nearly three years.

🌊 The seawater thing

The university’s release is careful here. It says the steel performs comparably to titanium for producing hydrogen from “desalted seawater or acid.” Desalted. Meaning: you desalinate first, then electrolyse.

The paper is careful too. The tests were run in 3.5 wt% NaCl — simulated seawater, the standard laboratory solution for chloride pitting tests. Not ocean water. Not desalinated water. A salt solution in a beaker, which is the correct way to measure what they were measuring.

Somewhere downstream, “desalted seawater” became “seawater,” and then “seawater” became “directly from the ocean,” and one outlet arrived at steel “made from seawater,” which is not a sentence about anything.

Why this matters: direct seawater electrolysis is one of the most contested ideas in hydrogen research, and the economics are not close.

The canonical analysis — Hausmann, Schlögl, Menezes and Driess in Energy & Environmental Science, 2021 — puts it bluntly:

The numbers behind that: making a kilogram of hydrogen needs roughly 20 litres of water in practice. Desalinated water runs about $0.85 per cubic metre. That’s under two cents of desalination per kilogram of hydrogen — against an electricity bill of two to three dollars per kilogram. Desalination is a rounding error.

The same group’s 2024 follow-up in Joule is titled, with no ambiguity at all, “Hyping direct seawater electrolysis hinders electrolyzer development.” A 2024 European Commission review of 748 papers found no project demonstrating a clear benefit of direct over indirect seawater electrolysis.

In fairness, there is a real counter-argument: space, not money. An offshore platform has nowhere to put a desalination plant, and the smallest electrolysers need so little water that the economics of a tiny desalination unit look worse than a big one. That’s a legitimate niche and serious people work on it.

But “this steel makes hydrogen straight from the sea” is not a description of a niche. It’s a description of something that isn’t happening.

🧪 The mismatch that nobody seems to have noticed

This is the part I found most interesting, and I want to be precise about it because it’s a criticism of a claim, not of a material.

SS-H2’s demonstrated superpower is resisting chloride at high voltage. That’s what 3.5% NaCl tests. That’s what the 1700 mV figure means. Chloride is the aggressor; the manganese layer is the defence.

The application in the headline is a PEM electrolyser.

A PEM electrolyser runs on ultrapure water. Not low-chloride water — ultrapure. One major manufacturer’s specification calls for deionised water at a minimum of ASTM Type II, above 1 megohm-centimetre, with Type I preferred. Chloride in that context isn’t the operating environment; it’s a contaminant, and a notorious one. Published work suggests damage thresholds in the region of ten parts per million.

The laboratory test solution was around 21,000 ppm chloride. The machine’s tolerance is around 10 ppm. Those are different worlds by three orders of magnitude.

So the steel’s headline advantage — chloride resistance — is an advantage against something a PEM electrolyser is specifically engineered never to contain.

To be scrupulously fair, this does not make the work irrelevant to electrolysers, for two reasons:

• The voltage problem is real regardless. Stainless steel’s transpassive breakdown near 1000 mV happens in sulfuric acid too. A steel that stays passive above the water-oxidation potential is a genuinely interesting proposition for acid service, chloride or no chloride.

• The group knows this. More on that below — they went and did the acid experiment.

But there are two failure modes that killed coated stainless steel in PEM before, and neither appears in the 2023 paper. The first is contact resistance: oxide films are insulators, which is exactly why titanium needs a precious-metal coating — not for corrosion protection, but for conductivity. SS-H2’s whole mechanism involves growing a second oxide on top of the first. The second is ion release: dissolved iron and chromium from steel are documented membrane poisons, and manganese ions are known to migrate in this class of membrane.

The 2023 paper measures neither. Its reference list contains no membrane literature, no bipolar-plate literature, and no electrolyser durability literature — because it is a corrosion paper, and a good one, that was press-released as an energy-economics story.

In three years and roughly sixty articles, I could not find a single electrochemist publicly checking the 40× claim. Not a rebuttal, not a caveat, not a blog post. The specialist hydrogen trade press — the outlets that would have caught this — never ran the story at all.

🔭 The genuinely new part, which nobody covered

Here’s the irony, and it’s a good one.

While the 2023 press release was being recycled for the fifth time, the same group quietly published three follow-up papers — and they are addressing precisely the objections above. None of them got a headline.

The 2025 acid paper reports stable operation for over 100 hours with 99.9% faradaic efficiency. That’s real, hard-won data, and it’s the first durability number in this whole story.

It also needs reading with a ruler. It was run at 10 milliamps per square centimetre. Industrial PEM electrolysers operate at 1–2 amps per square centimetre — one to two hundred times higher. And an industrial stack is expected to last more than 60,000 hours. So: 100 hours at 1% of operating current, against a target of 60,000 hours at full current.

That is not a criticism. That is what honest early-stage research looks like, and reporting it accurately is more interesting than another round of “stuns scientists.” The group is doing exactly the right work in exactly the right order. The distribution layer just isn’t interested in the part where you find out whether it works.

🛠️ How not to get caught by this

None of the above requires you to be a metallurgist. It requires four habits, and they generalise to every technology story you’ll read this year.

1. Scroll to the journal citation before you get excited. ScienceDaily, EurekAlert and most aggregators print the year in the reference block at the foot of the article. If the body text has no date and the citation says 2023, you are reading an old story in new clothes.

2. Ask what the percentage is a percentage of. “40× cheaper” was always about one material line item. The three boundary-crossings in the table above — material → part, stack → plant, capex → product cost — are the most common way a true number becomes a false impression.

3. Check whether the test conditions match the advertised application. Chloride resistance sold into a machine that runs on ultrapure water is the tell here. Ask: what did they actually put it in, for how long, at what load?

4. Look for durability, not peak performance. A breakdown potential from a single voltage sweep is a ceiling, not a lifetime. The question is always hours × current density, and if a story doesn’t give you both, it hasn’t told you whether the thing works.

And two for anyone on the other side of it:

5. Press offices: put the publication date in the body text. Not the footer. One clause — “published in 2023” — would have stopped this entire cycle five times over.

6. If you must attach an economic estimate to a science result, cite it. The HK$17.8m and the 53% have been repeated across three years and sixty-odd outlets with no source attached to either. Someone, somewhere, calculated them. Nobody can check them.

🧷 What I keep thinking about

The steel is real. The manganese thing is genuinely strange and genuinely important, and a PhD student refusing to believe his own data until the atomic-scale evidence piled up is science working exactly as advertised.

Nothing in this issue takes any of that away.

What’s broken is downstream. A careful piece of corrosion research acquired a cost estimate at the communications stage, and that estimate — untraceable, unchecked, about the wrong thing — became the headline. Then the headline outlived the news, was rewritten as new five times, and lost its date somewhere on the fourth pass.

Meanwhile the same team published the follow-up experiments that would tell you whether any of it works. Three papers. A hundred hours of real durability data. A different mechanism using nickel. Not one headline between them.

The system rewarded the press release and ignored the science.

That’s not a story about steel. It’s the reason you should read the small type at the bottom of anything that stuns scientists.

📚 Sources

Every claim in this issue was checked against the primary source before publication. Where the university’s press release and the published paper describe something differently, both are quoted rather than reconciled. Two figures I could not verify — the alloy’s exact manganese content, and the “two patents granted” claim, which has no public patent number — are named here rather than repeated as fact.