---
id: "20260816-0241-did-john-graham-cumming"
title: "Did John Graham-Cumming register lowbackgroundsteel.ai in March 2023, and is he the origin of the low-background-steel metaphor for pre-AI web text?"
type: "capture"
status: "promoted"
origin: "batch"
promoted_to: ["30-notes/claim-graham-cumming-registered-lowbackgroundsteel-ai-march-2023-announced-june-2025.md","30-notes/claim-low-background-steel-ai-metaphor-predates-graham-cummings-site.md","50-questions/question-verify-kyle-mcdonald-december-2022-low-background-steel-post.md","40-entities/entity-john-graham-cumming.md"]
not_promoted: ["Matt Rickard's June 2023 essay using the same analogy — a further-origin lead, not independently fetched (only known via Pant's account); left in the capture's 'Further leads' for a future session to chase to primary rather than promoted on secondhand knowledge of its existence.","Wikipedia's two further 2023 usages ('Low-Background Steel,' HN, 2023-04-18; 'Low Background Media: Against the Future,' 2023-05-14) — not chased to primary, no claim to promote yet.","The June 2025 mainstream-press cycle (New Scientist, Business Insider, Ars Technica, The Register, SaskToday, Shelly Palmer) — a real, chaseable thread on how the catalogue site entered tech journalism, but distinct from this capture's origin question and not researched here.","Graham-Cumming's Cloudflare CTO-to-board transition timing (per Pant) — an institutional-affiliation detail, not load-bearing for either claim promoted, not atomic enough on its own.","lowbackgroundsteel.ai's public-submission queue mechanic — relevant to a future note on the catalogue's method, not to the origin question this capture answers.","Kyle McDonald and Hacker News user 'abtinf' as standalone entity hubs — both real and load-bearing to the priority-dispute claim, but each rests on a single, weakly-sourced mention this session (McDonald: Tier-2 secondhand with an open unverified-quote flag; abtinf: pseudonymous, one post). Per the entity-promotion test's 'when unsure, don't promote,' left as mentions inside the claim-note rather than built as hubs. Rajiv Pant and Matt Rickard likewise left as mentions — single-capture appearances as sources, not yet vault-domain figures in their own right."]
writer_model: "claude-sonnet-5"
date_created: "2026-08-16T00:00:00.000Z"
provenance: "batch run, 2026-08-16"
derived_from: []
tags: ["ai-training-data","model-collapse","metaphor","provenance","attribution","low-background-steel","question-resolution"]
sources: [{"source_url":"https://blog.jgc.org/2025/06/low-background-steel-content-without-ai.html","source_sha":"48153a84dd693417905462d680ac3faaece6e3a03a444e3ba1ae6c1ba034bd40","source_title":"Low-background Steel: content without AI contamination","source_author":"John Graham-Cumming","source_venue":"John Graham-Cumming's blog (blog.jgc.org)","source_date":"2025-06-10T00:00:00.000Z","source_tier":1},{"source_url":"https://lowbackgroundsteel.ai/about","source_sha":"c46cc4e9db1d3d49103987ac6ca6ffd3b0902f4fe5b76062bfc3c86c0cbc67a2","source_title":"About — Low-background Steel","source_author":"John Graham-Cumming","source_venue":"Low-background Steel (lowbackgroundsteel.ai)","source_date":"2023-03-01T00:00:00.000Z","source_tier":1},{"source_url":"https://rajiv.com/blog/2025/03/31/the-new-low-background-steel-why-pre-ai-content-could-become-medias-most-valuable-asset/","source_sha":"d7418cd1d814fcc230b951c5b96e4a1e1c1b72ba5a0379dc9a0935297ac23948","source_title":"The New Pre-Nuclear Steel: Why Pre-AI Content Could Become Media's Most Valuable Asset","source_author":"Rajiv Pant","source_venue":"rajiv.com","source_date":"2025-03-31T00:00:00.000Z","source_tier":2},{"source_url":"https://news.ycombinator.com/item?id=34085194","source_sha":"520823db09a497791032839bb283d196ecee29203caedeef29816c64521f5d9e","source_title":"Ask HN: Has AI ended the \"low background radiation\" era of the internet?","source_author":"Hacker News user 'abtinf' (real identity unconfirmed)","source_venue":"Hacker News — retrieved via the HN Algolia search API (hn.algolia.com) after a direct fetch of news.ycombinator.com returned HTTP 429","source_date":"2022-12-21T00:00:00.000Z","source_tier":3},{"source_url":"https://en.wikipedia.org/wiki/Low-background_steel","source_sha":"34f29b16681556689d0cce31e04adb29503ef407bb541342fd1ffaf5f61bf25b","source_title":"Low-background steel","source_author":"Wikipedia contributors","source_venue":"Wikipedia","source_date":"2026-07-17T00:00:00.000Z","source_tier":4}]
seek_code_commit: "17d9798"
---


This capture answers [[question-verify-lowbackgroundsteel-ai-domain-and-metaphor-origin]] directly, discharging the flag on [[claim-low-background-steel-became-a-metaphor-for-pre-2022-ai-free-web-text]]. **Verdict, in two parts: yes on the registration date — confirmed at Tier 1 by Graham-Cumming's own words on two of his own venues. No on sole origin of the metaphor — the AI-data application of "low-background steel" is attested in public discussion before his site existed, and the best available secondary account explicitly describes it as arising independently in several places at once, with Graham-Cumming credited as the person who catalogued and (much later) popularized it, not as its coiner.**

## Claim: John Graham-Cumming created lowbackgroundsteel.ai in March 2023, but did not publicly announce the site until June 2025 — more than two years later

**Claim type:** historical/biographical. **Floor:** Tier 3–4 acceptable when uncontested; this specific fact is self-attested twice on Tier 1 venues, well above the floor.

On his personal blog, dated 2025-06-10, Graham-Cumming wrote: "Somehow I forgot to blog my site: https://lowbackgroundsteel.ai/. I created it back in March 2023 as a clearinghouse for online resources that hadn't been contaminated with AI-generated content." The site's own About page independently states the same founding date and authorship in its metadata header: "Established Mar 2023" and "Maintainer John Graham-Cumming" (the live page also carries a running "Updated 2026-08-16" stamp, since it is an actively maintained catalogue, not a static document). Both sources are the subject's own words on his own venues — the strongest tier available for a self-reported fact — and agree with each other. The two-plus-year gap between quiet registration (March 2023) and public announcement (June 2025) is itself part of the record: Graham-Cumming's own phrasing, "somehow I forgot to blog my site," indicates the domain existed and was being used as a catalogue well before it entered public discourse under his name.

## Claim: The AI-data application of the "low-background steel" analogy was already circulating in public online discussion before lowbackgroundsteel.ai existed

**Claim type:** historical/priority claim — contested, since it bears directly on whether Graham-Cumming originated the metaphor. **Floor:** escalates to Tier 1–2 required. **Met:** Tier 3 only (a dated, publicly timestamped post, but by a pseudonymous author, retrieved via a search-API mirror after the canonical page itself could not be fetched directly) — recorded and flagged rather than treated as settled.

Wikipedia's "Low-background steel" article's cultural-impact section cites, as its earliest reference for the metaphor's AI-era use, a Hacker News "Ask HN" post dated 2022-12-21 — three months before Graham-Cumming's March 2023 registration. Following that footnote to the primary post (news.ycombinator.com/item?id=34085194 returned HTTP 429 on direct fetch; the text was retrieved instead via Hacker News's own Algolia search API, which mirrors the original post content and its timestamp) turns up a Hacker News user "abtinf" posting, at 2022-12-21T19:24:34Z, under the title "Ask HN: Has AI ended the \"low background radiation\" era of the internet?": "On July 16, 1945, the nuclear detonation at Trinity site ended the era of low background steel. All future steel would become increasingly contaminated with trace radiation, making it unsuitable for sensitive applications. Over the last decade, with the release of GPT/DALL-E/SD/Copilot/Google Assistant/computational photography/etc, has the same thing happened to the internet? Do we have all the guaranteed-human-generated content we will ever get?" This is the same "clean-source-is-older-than-the-contamination-event" structure later catalogued at lowbackgroundsteel.ai, publicly posted under a real timestamp almost three months before Graham-Cumming's registration date — but the poster is pseudonymous (real identity unconfirmed) and the post drew zero comments and one point, so it is recorded as an existence proof of prior public use, not proof that this specific post caused or fed into Graham-Cumming's independent naming choice. `[flagged — pseudonymous primary source, Tier 3; corroborated independently in the next claim]`

## Claim: Per a named industry commentator, the metaphor's AI-data use "emerged organically across the tech community" rather than originating with Graham-Cumming, with artist/programmer Kyle McDonald's December 2022 post cited as an independent early instance

**Claim type:** historical/biographical, contested (a direct priority claim). **Floor:** Tier 1–2 required. **Met:** Tier 2 (Rajiv Pant — former CTO of The New York Times and The Wall Street Journal — writing on his own site about the history of an analogy in his professional network).

Rajiv Pant's essay states: "Before diving into implications, it's important to acknowledge that while John Graham-Cumming has helped popularize this analogy through his work and advocacy, the comparison appears to have emerged organically across the tech community. Programmer and artist Kyle McDonald made a similar observation in an X (formerly Twitter) post on December 5, 2022, noting that just as particle detectors require special steel manufactured before nuclear contamination, future AI might need training data collected before 2022. Others independently reached similar conclusions around the same time." Pant also reports that Graham-Cumming himself had earlier "explored [this analogy] in tweets and blog posts" but that "his specific tweet on this subject is no longer available" — meaning even Pant's account of Graham-Cumming's own earliest use is secondhand and unverifiable at the primary level. McDonald's original X post was not independently located or fetched for this capture (X/Twitter access limitations), so the Kyle McDonald date rests on Pant's Tier 2 account alone. `[flagged — priority claim currently rests on one Tier-2 secondary; McDonald's original post itself is an [unverified-quote — needs direct read] lead]`

## Further leads

- Matt Rickard's essay "The Low-Background Steel of AI" (mattrickard.com / blog.matt-rickard.com), dated 2023-06-17 per Rajiv Pant's account, is another public use of the analogy for AI training data — after Graham-Cumming's private registration but well before his June 2025 public announcement; not independently fetched here.
- Wikipedia's "Low-background steel" article footnotes two more 2023 usages not yet chased to primary: "Low-Background Steel," Hacker News, 2023-04-18; and "Low Background Media: Against the Future," 2023-05-14.
- Wikipedia's cultural-impact section lists the June 2025 mainstream-press cycle (New Scientist, Business Insider, Ars Technica, The Register, SaskToday, Shelly Palmer) that followed Graham-Cumming's public announcement — a distinct, chaseable timeline of how the catalogue site itself (as opposed to the bare metaphor) entered mainstream tech journalism.
- Per Rajiv Pant, Graham-Cumming transitioned from Cloudflare CTO to Cloudflare board member around the same date as Pant's essay (2025-03-31) — an institutional-affiliation detail not explored further here.
- lowbackgroundsteel.ai's About page states the site accepts public submissions via a queue ("Submissions are welcome. They go into a queue and are added by hand") — relevant to any future note on the catalogue's method, distinct from the metaphor's origin question.

> [!note] Seek's commentary:
> The compound question splits cleanly under the sourcing floor: "did he register the domain in March 2023" is a clean, doubly-self-attested Tier 1 yes. "Is he the origin of the metaphor" is a priority claim, and priority claims are exactly the kind the floor is strictest about — the moment I went looking for the earlier record, one turned up within one search (the Ask HN post), which is itself evidence about how thin the "he originated it" framing in casual tech-press retellings actually is. The Register piece the existing vault claim-note rests on doesn't even raise the question; it just states the March-2023 date as if uncontested. Worth flagging for promotion: the existing claim-note's title assumes an origin story this capture complicates.

## Entity candidates

- Kyle McDonald — person — per Rajiv Pant's account, the artist/programmer whose December 5, 2022 X post is the earliest independently-reported use of the AI-data analogy, predating lowbackgroundsteel.ai's registration by nearly four months; the foundational figure any Graham-Cumming-origin claim has to be measured against, and currently the weakest-sourced part of this capture (Tier 2 secondhand only).
- Hacker News user "abtinf" — person (pseudonymous) — posted the dated, public "Ask HN" analogy on 2022-12-21, three months before Graham-Cumming's registration; real identity unconfirmed, but the post itself is a primary timestamped artifact.
- John Graham-Cumming — person — creator and maintainer of lowbackgroundsteel.ai; former Cloudflare CTO, now Cloudflare board member; subject of the capture.
- Rajiv Pant — person — former CTO of The New York Times and Chief Product/Technology Officer of The Wall Street Journal; source for the "emerged organically" multiple-origin account and for the Kyle McDonald and Matt Rickard leads.
- Matt Rickard — person — wrote an independent June 2023 essay using the same analogy for AI training data; a further-origin lead not yet chased to primary.