AnomalyRegistry
AR-0084

The Oklo Natural Reactor

Later explained
Date Operated approximately 1.7 billion years ago. Discovered June 1972.
Location Oklo uranium mine, Haut-Ogooué Province, Gabon
Coordinates -1.4000, 13.1500

Summary

In 1972 a French technician found that a batch of uranium ore was missing some of its uranium-235. The shortfall was tiny, three parts in a thousand, but in nuclear materials accounting a missing gram of fissile material is a serious matter, and it triggered an investigation.

The uranium had not been stolen. It had been used. Roughly 1.7 billion years ago, in a seam of rock in what is now Gabon, uranium ore went critical and ran as a nuclear reactor, on and off, for a few hundred thousand years, with nobody present to build it, fuel it, or moderate it.

why it could happen then, and can never happen again MODERN REACTOR FUEL (~3%) OKLO RUNS 1.7 bn years ago 0.72% today 2.0 bn 1.0 bn now 3% 0% groundwater moderated it. when it boiled off, the reaction stopped, and restarted when the water returned. ~30 min on, ~2.5 hrs off. for a few hundred thousand years. no one built it.
Figure Drawn by Anomaly Registry. Uranium-235 has a half-life of about 704 million years against uranium-238's 4.5 billion, so natural uranium was around 3 per cent U-235 when Oklo ran, roughly the enrichment of modern reactor fuel. It is 0.72 per cent today. The window is closed permanently.

What is documented

The discrepancy. In June 1972, routine mass spectrometry at the Pierrelatte enrichment plant in France showed uranium hexafluoride with a uranium-235 concentration of about 0.717 per cent. The figure for natural uranium, anywhere on Earth, in any ore, from any planet in the solar system, is about 0.720 per cent. The difference is around three thousandths of a per cent.

It should not have been possible. Uranium isotope ratios are fixed by the age of the solar system and are the same everywhere. A shortfall meant either an instrument fault or a diversion of fissile material.

The trace-back. The ore was traced to the Oklo mine in Gabon. Further sampling there found ore far more depleted than the original batch, in places down to around 0.44 per cent uranium-235. Something had consumed a substantial fraction of it.

The fission products. The decisive evidence was not the missing uranium but what had replaced it. The rock contained the characteristic isotopic signature of nuclear fission: neodymium, ruthenium and other elements present in the exact proportions that fission produces and in proportions found nowhere in ordinary rock. Xenon isotopes trapped in the mineral grains matched fission yields.

The ore had not merely lost uranium. It contained the ash.

The reactor zones. Sixteen or more distinct natural reactor zones have since been identified at Oklo and the nearby sites of Okelobondo and Bangombé.

How it ran. Reconstruction from the isotopic record indicates that the zones operated intermittently, in cycles, over a period of the order of hundreds of thousands of years, at an average thermal power of roughly a hundred kilowatts. This is a very small reactor. It is about the output of a domestic heating system, sustained for longer than modern humans have existed.

Leading explanations

The mechanism is understood, and it turns on two facts that were true then and are not true now.

The fuel was enriched, because the universe was younger. Uranium-235 decays faster than uranium-238. Its half-life is about 704 million years; uranium-238's is about 4.5 billion. Run the clock back 1.7 billion years and the ratio was very different: natural uranium at Oklo would have been around three per cent uranium-235.

Three per cent is not a coincidence. It is approximately the enrichment level of the fuel in a modern commercial power reactor. The Earth was, for a time, made of reactor fuel, and it has been decaying out of that condition ever since. Oklo could happen then. It cannot happen now, anywhere, because the fuel no longer exists.

Water was the moderator. A chain reaction needs the neutrons slowed down, and water does this. Groundwater percolating through the porous, uranium-rich sandstone at Oklo did the job that a modern reactor does with a deliberately engineered moderator loop.

It regulated itself, which is the best part. As the reaction ran, it heated the water. The water boiled away. With the moderator gone, the neutrons were no longer slowed, the chain reaction stopped, and the rock cooled. Groundwater seeped back in, and the reaction restarted.

The isotopic evidence is consistent with a cycle of roughly half an hour of criticality followed by a couple of hours of dormancy while the water returned, repeating for geological ages. Nobody built the control system. The physics supplied one.

Why nobody had predicted it. Somebody had. Paul Kuroda published the conditions required for a natural fission reactor in 1956, sixteen years before one was found, and the conditions he specified are essentially the ones that obtained at Oklo. His paper was a curiosity until it stopped being one.

What the popular version gets wrong

"A two-billion-year-old nuclear reactor: proof of ancient technology." It is the opposite of proof of technology. Its entire significance is that no technology was involved. Oklo is remarkable precisely because uranium, water and time did unassisted what it took the Manhattan Project to arrange deliberately.

"It exploded." It did not, and it could not. A reactor is not a bomb; sustaining a slow chain reaction and producing a nuclear detonation are different physical problems, and the second one is very hard. Oklo simmered at about a hundred kilowatts. It was, by any standard, an extremely boring reactor.

"Scientists cannot explain how it started." They can. The isotope ratios of the era, the porosity of the sandstone, the presence of groundwater, and Kuroda's 1956 conditions account for it completely.

"It could happen again." It cannot. Natural uranium is now about 0.72 per cent uranium-235 and will never again be three per cent. The window closed permanently, roughly a billion years ago, and the last natural reactor on Earth has already run.

Current status

Later explained. The registry keeps this record as its cleanest case of an explanation that is more astonishing than the anomaly it replaced.

And it turned out to be useful. Because the reactors at Oklo produced highly radioactive fission products, and because a great many of those products stayed more or less where they were formed for 1.7 billion years, Oklo is now studied as a natural experiment in exactly the problem that no one has solved: whether nuclear waste can be safely left in rock for geological time. The site has been running that experiment, unattended, since before multicellular life.

Sources

  • Perrin, F., and Commissariat à l'énergie atomique (1972). Initial identification of the uranium-235 depletion at Pierrelatte and trace-back to Oklo.
  • Kuroda, P. K. (1956). Theoretical conditions for the occurrence of a natural fission reactor, published sixteen years before one was found.
  • International Atomic Energy Agency, proceedings on the Oklo phenomenon (Libreville, 1975 and later).
  • Isotopic analyses of fission-product neodymium, ruthenium and xenon in the Oklo reactor zones.
  • Subsequent literature on Oklo as a natural analogue for deep geological disposal of nuclear waste, and on constraints it places on variation in the fine-structure constant.

Last reviewed: July 2026. Records are provisional. Where the evidence changes, the entry changes. Found an error? Tell us.

Advertise on Anomaly Registry

A documentary archive of 100 investigated cases, read slowly and at length by an audience that arrives on purpose. Below are the current delivery figures and what is available to buy.

Figures are rolling daily averages, current as of September 2026. They are restated as the traffic changes rather than left to age quietly, and we are happy to screen-share the underlying Ad Manager and analytics reporting before an insertion order is signed.

Anomaly Registry is a reference archive rather than a feed. Every one of its hundred records is written to a fixed structure, sourced to contemporaneous documents, inquest files, peer-reviewed literature or credible reporting, and each carries a section setting out precisely what the popular version of the story gets wrong. Cases that were later explained and cases that were comprehensively debunked are kept alongside the unsolved ones, because an archive that keeps only its mysteries is not an archive. That editorial position is the reason people stay.

It also explains the shape of the traffic. A session here runs to more than four pages and close to nine minutes, which is several times what a typical content site sees. Readers arrive looking for one case, find the sources laid out, and follow the cross-references into the next record. For an advertiser, the practical consequence is that a banner on this site is in front of a settled, attentive reader for minutes rather than seconds, and it is seen repeatedly across a single visit without ever being the reason someone leaves.

Three placements are available on every page. The leaderboard sits above the masthead and accepts 728x90, 970x90 and 320x50. The in-article unit sits inside the body of the record, where the reader is already stopped, and accepts 728x90, 300x250 and 336x280. The footer unit closes the page and accepts 728x90, 970x90 and 320x50. There are three slots per page and there will not be a fourth: the inventory is deliberately limited so that each impression is worth more than it would be on a page carrying eight.

What we do not run matters as much as what we do. There are no interstitials, no pop-ups, no auto-playing video, no content-recommendation chumboxes, and nothing that reflows the page after the reader has started reading. Ads are served in reserved space so the layout does not jump. This protects viewability for the advertiser as directly as it protects the reading experience, and it is the reason we can quote the session figures above with a straight face.

The editorial environment is unusually safe for a subject area that is often anything but. This site does not publish conspiracy content, it does not promote pseudoscience, and it does not treat a rumour as a finding. Its entire method is to separate what is documented from what has merely been proposed, and to say plainly when a famous mystery has an unglamorous answer. A brand placed here sits next to sourced, corrected, carefully hedged writing rather than next to speculation dressed as news.

The readership is a general-interest one with a research habit: people who read long-form nonfiction, follow science and history coverage, and check citations. In practice the categories that fit well are books and publishing, documentary and streaming, museums and exhibitions, education and online courses, audio and podcasts, consumer technology, outdoor and travel, and financial services. If your product rewards attention and explanation rather than impulse, this is a good room to be in.

Everything is delivered through Google Ad Manager, so you can buy the way you already buy. Direct-sold campaigns can be trafficked as a guaranteed line item with standard reporting on impressions, viewability and click-through. Programmatic demand can reach the same inventory through the open auction or through a preferred deal or private marketplace at a floor we agree in advance. We accept standard IAB display creative and HTML5, and we can accommodate a reasonable third-party ad server or verification tag.

Category sponsorship is available for advertisers who want something more durable than a rotation. The register is organised into seven categories, and a sponsor can take the whole of one, appearing across every record inside it with an acknowledgement in the page furniture. Sponsorship never buys editorial influence: a sponsor cannot commission a record, cannot review one before publication, cannot have a record altered, and cannot have one removed. That rule is not negotiable and it is the thing that keeps the inventory worth buying.

Some categories are declined outright, and it is fairer to say so up front than to discover it after an insertion order. We do not accept adult content, gambling, payday or high-cost short-term credit, cryptocurrency speculation and token sales, dietary supplements making health claims, political advocacy, psychic or clairvoyant services, or anything that reads as a scare advertisement dressed up as an article. Nothing may be styled to look like a record or like part of the editorial page.

Rates depend on placement, volume, whether the buy is run-of-site or targeted to a category, and how long it runs. There is no rate card on this page because a sensible number depends on what you are trying to do, and quoting one in the abstract usually wastes both parties' time. Tell us the campaign, the flight dates and the budget you are working to, and you will get a straight answer including whether we think the fit is wrong.

To start, write to or use the contact form and mark the message for the advertising desk. Include the flight dates, the placements you want, the creative sizes you have ready and any verification requirements, and we will come back with availability and a quote. Media buyers who want the underlying analytics before committing are welcome to ask; we would rather show the reporting than be taken on trust.

Privacy Policy

This policy explains what information Anomaly Registry collects when you visit this website, why it is collected, how it is used, and what choices you have. It applies to anomalyregistry.com and to any subdomain operated by us. It was last updated on the date shown at the end of this policy, and we will revise it if our practices change.

We have designed this site to collect as little personal information as possible. You do not need an account to read anything here. There is no login, no user profile, and no comment system. In the ordinary course of reading the register, you are not asked to identify yourself, and we do not ask you to.

Like virtually every web server, ours produces access logs. These logs record the internet protocol address of the requesting device, the date and time of the request, the page or file requested, the referring page if one was sent, and the browser user agent string. These logs exist for security, abuse prevention, and diagnosing technical faults. They are retained for a limited period and then discarded. We do not use them to build profiles of individual readers.

If you write to us through the contact form, we receive whatever you choose to put into it: typically a name, an email address, and a message. We use that information for the sole purpose of reading and answering your message. Writing to us does not add you to any list, and we do not sell, rent, or trade contact details with anyone.

There is one optional mailing list, and you only join it by asking. If you submit the subscribe form we store the address you gave, the time you submitted it, the IP address it came from, and the exact wording of the consent you agreed to. We keep that record because Canadian anti-spam law expects us to be able to show when and how you asked to hear from us. The list is used to say that records have been published or materially revised, and for nothing else. Every message carries a way to leave, and asking to be removed by email works just as well.

Corrections are a special case, and worth stating plainly. If you write to point out an error in a record, we may act on that correction and revise the entry. We will not publish your name or your email address in connection with a correction unless you explicitly ask us to be credited.

This site is supported by advertising served through Google Ad Manager. Advertising technology providers may set and read cookies or similar identifiers on your device in order to select advertisements, limit how often you see the same advertisement, and measure performance. These identifiers are set by those third parties, not by us, and the data they collect is governed by their own privacy policies rather than this one.

We do not have access to the personal data that advertising partners collect, and we cannot delete it on your behalf. If you wish to limit interest-based advertising, you can adjust the advertising settings offered by Google, use your browser's cookie controls, or use the industry opt-out tools that are available in your region. Details are set out in our Cookie Policy.

We may use privacy-respecting aggregate analytics to understand which records are being read and how people move through the register. Where we do, we configure it to avoid collecting information that identifies individual readers. Aggregate figures tell us that a case page is popular; they do not tell us who read it.

We do not knowingly collect personal information from children. This site is a documentary archive intended for a general adult and older student readership. If you believe a child has sent us personal information through the contact form, write to us and we will delete it.

Depending on where you live, you may have rights over personal data relating to you, including the right to ask what we hold, to ask for a copy, to ask for correction, and to ask for deletion. Because we hold so little, these requests are usually simple to answer. Write to us and we will respond within the period required by the law that applies to you.

Our servers are located in North America, and information sent to this site is processed there. If you are visiting from elsewhere, your request necessarily crosses borders in order to reach us. By using the site you understand that this transfer takes place.

If we change this policy, the revised version will be posted here with a new date. Material changes will be noted plainly rather than buried. Questions about this policy should be sent through the contact page.

Cookie Policy

A cookie is a small text file that a website asks your browser to store, and that your browser sends back on later requests. Similar technologies, including local storage and pixel tags, do comparable work. This policy explains which of these are used on Anomaly Registry, by whom, and what you can do about them.

The registry itself is a reading site. It does not require a cookie to show you a record. We do not set cookies to track you across the internet, and we do not operate a first-party advertising identifier of our own.

Strictly necessary cookies may be set to keep the site secure and functioning, for example to help mitigate automated abuse of the contact form. These are limited in scope and are not used for advertising or profiling.

Advertising cookies are set by Google Ad Manager and by the advertising technology partners that operate within it. Their purposes include selecting which advertisement to show, capping the number of times a given advertisement is repeated, detecting invalid traffic and click fraud, and measuring whether an advertisement was actually seen.

Some of these advertising cookies support interest-based advertising, in which the advertisement you see is influenced by inferences drawn from browsing activity over time and across sites. Whether this occurs depends on your region, your settings, and the choices you have already made with those providers.

We may use aggregate analytics to count page views and understand which records are read. Where an analytics tool is used, we prefer configurations that do not identify individual readers and that do not follow readers across other websites.

We do not embed social media widgets, share buttons, or tracking pixels from social networks. This is a deliberate choice for an archive of this kind, and it removes an entire category of third-party tracking that most content sites carry.

Your browser gives you direct control over cookies. Every major browser can block third-party cookies, delete existing cookies, and clear stored data on exit. Blocking third-party cookies will not prevent you from reading anything on this site, though it may make the advertising you see less relevant.

You can also manage advertising preferences with Google directly through its advertising settings, and through the industry opt-out mechanisms operated in various regions. Because these are operated by third parties, we cannot make the choice on your behalf.

Where the law in your region requires consent before non-essential cookies are set, a consent notice will be shown and your choice will be respected. Withdrawing consent later is possible through the same mechanism.

Advertising partners may change over time. Rather than publish a list that goes stale, we identify the advertising system in use, which is Google Ad Manager, and direct you to Google's own disclosures for the current set of partners operating within it.

If you have a question about cookies on this site that this policy does not answer, write to us through the contact page and we will answer it.

Anomaly Registry is an independent documentary archive. It is published as a work of reference and public interest reporting, and it is supported by advertising. It is not affiliated with, endorsed by, or acting on behalf of any government, university, police service, coroner, scientific institution, or investigative body named in any record.

The register documents real events, including in some cases deaths, criminal investigations, and inquests. Where a record concerns a criminal matter, nothing on this site should be read as an allegation of guilt against any living person. Where an individual has been named in the public record, we report the fact of that naming and its evidentiary standing; we do not accuse.

Records are written from sources. Where a source is a primary document, we say so. Where a claim rests on secondary reporting, we say so. Where sources conflict, we state the conflict rather than quietly choosing a side. Where a widely repeated claim cannot be traced to any source, we say that too, because an untraceable claim is itself a finding.

Every record separates what is documented from what has been proposed. Hypotheses are labelled as hypotheses and attributed to whoever advanced them, together with their standing: peer-reviewed, contested, dismissed, or untested. We do not advance theories of our own.

The registry treats its own entries as provisional. Records carry a review date, and they are revised when the evidence changes. A record you read today may be corrected tomorrow, and that is a feature of the method rather than a defect in it.

Despite this method, errors will occur. If you find one, write to us. Corrections supported by evidence are made promptly, and we would rather be corrected than be consistent.

Scientific and forensic conclusions reported here are the conclusions of the researchers and officials who reached them, not ours. Reporting that a study proposed a mechanism is not an endorsement that the mechanism is correct, and the status assigned to a case reflects the state of the evidence rather than our opinion of it.

Nothing on this site is legal, medical, psychological, financial, or safety advice. Records that touch on medical or psychological matters, including mass psychogenic illness, report the diagnoses reached by clinicians at the time and afterward. They are not a substitute for professional care.

Quotations from sources are used sparingly and for the purposes of reporting, comment, criticism, and review. Copyright in quoted material remains with its owner. If you hold rights in material you believe has been used improperly, contact us and we will address it.

Original text, structure, record numbering, and editorial apparatus on this site are the property of Anomaly Registry. You are welcome to quote briefly and link to us. Wholesale reproduction of records, including reproduction for the purpose of training generative models, is not permitted without written permission.

Outbound links are provided as references. We do not control linked sites and are not responsible for their content, their accuracy, or their handling of your data.

This site is operated from British Columbia, Canada, and these notices are governed by the laws applicable there, without regard to conflict-of-law principles.

Accessibility Statement

Anomaly Registry is built to be usable by as many people as possible, including people who use screen readers, keyboard navigation, magnification, or other assistive technology. We treat accessibility as part of the build rather than as a later remediation.

We aim to meet the Web Content Accessibility Guidelines version 2.1 at level AA. This is a target we work toward continuously rather than a certification we claim to have completed.

The site is written as semantic HTML. Headings describe the real structure of a record, lists are marked up as lists, and the register is a genuine table with proper header cells, so that a screen reader can announce which column a value belongs to.

Every interactive element is reachable and operable by keyboard alone. Focus is visible at all times, and a skip link at the top of each page lets keyboard and screen reader users jump straight to the register without traversing the navigation.

Status is never conveyed by colour alone. Each status carries its written label in text, so a reader who does not perceive the colour difference between "Unsolved" and "Debunked" still receives the distinction.

Text colours have been chosen to meet or exceed the contrast ratios required at level AA against the page background. The typefaces are set at a comfortable reading size with generous line height, and the layout reflows to a single column on small screens without loss of content or function.

The site contains one animated element, an annotation on the home page. It respects the reduced-motion preference set in your operating system, and when that preference is on, the annotation appears without movement.

Text can be resized up to two hundred percent in the browser without breaking the layout or clipping content. Nothing on this site depends on a specific screen orientation.

Advertising is served by a third party. We control where advertisements are placed and label them as advertisements, but we do not control their internal markup, and their accessibility is ultimately determined by the advertiser.

We are developing an audio version of each record. When it is published, it will be an alternative way to receive the same content, not a replacement for the written entry, and the written entry will remain the authoritative version.

Accessibility work is never finished, and parts of this site will fall short of the standard we have set. If you encounter a barrier, we want to know about it. Please tell us what page you were on, what you were trying to do, and what assistive technology you were using.

Write to us through the contact page. We aim to respond within five business days and to fix genuine barriers promptly.

Terms of Use

These terms govern your use of anomalyregistry.com. By using the site you accept them. If you do not accept them, please do not use the site.

Anomaly Registry is offered as a free documentary reference. There is no account, no subscription, and no paywall, and nothing here is offered as a professional service of any kind.

You may read, quote briefly with attribution, link to, and print records for personal, educational, and research use. This is an archive, and it is meant to be used.

You may not reproduce records wholesale, republish the register or substantial parts of it, or scrape the site systematically. In particular, you may not reproduce this content for the purpose of training generative models. Permission for other uses can be requested through the contact page and is often granted.

Original text, the record numbering system, the status taxonomy, the editorial apparatus, and the design of this site are the property of Anomaly Registry. Third-party material quoted or referenced remains the property of its owners.

We work hard to make records accurate, and we correct them when they are not. Even so, the site is provided as it is, without warranty of any kind, express or implied, including any warranty of accuracy, completeness, or fitness for a particular purpose.

Records are provisional by design. A status can change, an identification can be confirmed or overturned, and a leading explanation can be displaced by a better one. Do not treat any record as a final word, and check the review date.

To the fullest extent permitted by law, Anomaly Registry and its operator are not liable for any loss or damage arising from your use of, or reliance on, anything published here.

If you send us a correction, a source, or a suggestion, you grant us permission to use it in the register. You will not be identified in connection with it unless you ask to be credited.

Do not use the contact form to send abuse, threats, spam, or automated submissions. We may block access to anyone who misuses the site or attempts to interfere with its operation.

Outbound links are provided as references and are not endorsements. We are not responsible for the content or conduct of any site we link to.

We may revise these terms. The version published here is the version in force. These terms are governed by the laws of British Columbia, Canada, and any dispute will be heard in the courts of that province.