Provenance Record

Analysis of how open-source licensing responds to AI-assisted rewrites, built from four cases (chardet LGPL to MIT to 0BSD with a Claude-assisted rewrite; the malus.sh cloning service; Modular's MAX License clause; Asahi Linux's move to the GPL), then cross-case implications: an unresolved legal question with named, differing tests (prior exposure vs structural similarity vs authorship), the same capability read as threat and as opportunity by named speakers, and different instruments binding different parties (a contract clause for MAX licensees, copyleft for distributors, permissive relicensing that drops obligations). Reported facts, the four separate developments and the 'What remains unknown' section are kept apart from the synthesis; the article states that none of the Modular and Asahi sources mention chardet and does not imply the cases are causally linked; no Herald prediction or opinion. 1,439 words (Analysis range 800-2000), 11 cited sources. Title is 147 characters (under the 150 limit).

Snapshots read from disk (gunzip + text extraction): source-0 (The Register, published 6 Mar 2026 01:00 UTC), source-1 (Ars Technica, Kyle Orland, published 2026-03-10T19:36Z), source-2 (PC Gamer, published 2026-03-31T11:18Z), source-4 (chardet issue 327), source-5 (issue 334), source-6 (PR 349), source-7 (release 7.0.0), source-8 (release 7.3.0), source-9 (Modular blog, datePublished 2026-10-22T16:07Z), source-10 (MAX_LICENSE on GitHub). The GitHub items were also cross-checked against the GitHub API (issue 327 created 2026-03-04T21:28:01Z, closed 2026-03-26T04:32:58Z with Blanchard citing Fontana's comment 4098524555; issue 334 created 2026-03-06, Fontana comment dated 2026-03-20; PR 349 opened 19:50:37Z and merged 19:53:37Z on 2026-03-22 by dan-blanchard; release 7.0.0 published 2026-03-04T00:11:12Z; release 7.3.0 published 2026-03-24T03:09:19Z). ROUTE FOR THE UNSNAPSHOTTED SOURCE: source-3 (Slashdot, 403 to the fetcher, manifest file null) was read as raw HTML by a live curl with a browser User-Agent (HTTP 200, 195,010 bytes, sha256 68a6cb5c5fb872db9517639bc38ff05473dee0195322ab86cc4336d72da92ac2), posted Monday 2026-11-09 12:35 AM by EditorDavid, 'It's FOSS reports:'; it confirms every Slashdot-attributed statement in the article (all repositories to GNU GPL version 2 or later; 'multiple attempts to exploit our permissive licensing'; 'in-development feature code was being lifted straight out of our personal forks and published without attribution or lazily rewritten with LLM assistance to obfuscate its origin'; 'made the difficult decision to begin the process of transitioning all of our repositories'; 'particularly from the BSD side'; OpenBSD relies on m1n1; Asahi forbids AI-generated contributions, which is It's FOSS's sentence relayed by Slashdot). The It's FOSS original returns a Cloudflare 'Just a moment' 403 page and Archive.org returned HTTP 429, so the It's FOSS text was not read; asahilinux.org/blog (HTTP 200) has no licensing post. Supplementary primary evidence read via the GitHub API and not cited by the article: AsahiLinux/asahi-audio commit 13ca3dd2be (2026-10-02T23:35Z, 'all: Relicence under the terms of the GNU GPL v2', 'Code checked in to asahi-audio hereafter shall be licenced under the terms of the GNU General Public License version 2'), AsahiLinux/speakersafetyd commit f447d9ae70 (2026-10-02T22:36Z, same wording), and AsahiLinux/m1n1 commit 2460b604c6 (2026-10-06T06:49Z: 'm1n1 has been relicensed under the terms of the GNU General Public License version 2' and the change 'only affects new code checked in from version 2.0.0 onwards'). They show the transition did begin, which supports 'Starts Moving' in the headline; they also show it began in early October 2026, earlier than the Slashdot post, and that it is not retroactive for m1n1. Allowlist: theregister.com, arstechnica.com, pcgamer.com, news.slashdot.org, github.com and modular.com are all already in config/source_allowlist.txt. Suspicious-pattern scan: no matches in any source. QUOTES: all 40 quoted strings in the body were machine-matched (normalised quotes) against the raw source text and all 40 matched; I also hand-checked speaker, source and context for each of the quotes in the headline-critical clauses and for the 20 others: Blanchard 'Claude gave me the ability to accomplish what I wanted in roughly five days' (Register, email); 'Anthropic's Claude is now listed as a project contributor' (Register narration); the poster 'Licensed code, when modified, must be released under the same LGPL license.' (issue 327 and Register); Blanchard 'extensive exposure to the original codebase' (issue 327); Blanchard 'the standard library code has to be a more lenient license like MIT or BSD' (issue 327, comment of 2026-03-05T14:09Z); JPlag 1.29% max for 7.0.0 vs 6.0.0.post1 (issue 327 table, labelled Blanchard's own); Kooyman (FSF executive director, by email) 'there is nothing 'clean' about a Large Language Model (LLM) which has ingested the code it is being asked to reimplement' and 'our strongest defense'; Perens 'The entire economics of software development are dead, gone, over, kaput!' and 'seems almost irrelevant' / 'may well be so too'; Ronacher 'heavily depends on copyrights and friction to enforce it'; Sanfilippo 'many people are disturbed' and 'build a new mental model and adapt'; Fontana 'I don't currently see any basis for concluding that chardet 7.0.0 is required to be released under the LGPL.' (issue 334, 2026-03-20, verbatim) and 'might be itself breaking new ground in open source' (issue 334 body, verbatim; the article's paraphrase of his question is accurate); PC Gamer 'a software architect at the UN Development Program' (verbatim, the outlet's spelling, correctly attributed to Mike Nolan; Dylan Ayrey 'founder of open-source software company Truffle Security'), 'legally distinct code with corporate-friendly licensing. No attribution. No copyleft. No problems.' (the site's wording per PC Gamer), 'a little tongue-in-cheek', 'AI clean-rooming'; Ars 'came to the forefront', 'Claude explicitly relied on some metadata files from previous versions of chardet', 'have yet to rule', 'an open question', 'still largely unsettled'; release notes 'Ground-up, MIT-licensed rewrite of chardet.', 'has been changed from MIT to', 'a maximally permissive license with no attribution requirement', 'should also be considered 0BSD licensed'; PR 349 'All chardet 7.x code is additionally licensed under 0BSD.' TITLE CLAUSES: (1) 'chardet Went From LGPL to MIT to 0BSD': LGPL to MIT in 7.0 on 2026-03-04 supported by The Register and Ars Technica and the 7.0.0 release notes; MIT to 0BSD supported by PR 349 and the 7.3.0 release notes (project's own pages; Register and Ars predate it, as the article says); Claude is named as the rewrite tool by The Register ('Anthropic's Claude is now listed as a project contributor'; Blanchard: 'Claude gave me the ability...') and by Ars Technica ('built with the help of Claude Code'). (2) 'Modular Bars AI Reimplementation of MAX': the blog says 'AI-assisted reimplementation of MAX: You can't use any part of MAX as input to an AI system to produce a reimplementation of MAX.' The MAX License (Last Modified October 21, 2026) section 1.4(d) reads: 'You shall not use MAX, or any portion of it, as training data, fine-tuning data, or input to any AI system in order to produce software that reimplements or substitutes for MAX, whether or not it contains literal copies of MAX. This does not restrict using AI tools to read, analyze, improve, or explain MAX, or to develop Your own software that runs on or interoperates with MAX.' It is a licence condition binding licensees, so 'bars' is not stronger than the clause (the clause is in fact broader than the article describes: see correction 2); the article correctly says it applies to licensees who agree to the terms, and calls MAX 'source available', not open source. (3) 'Asahi Linux Starts Moving to the GPL': supported by the Slashdot text read raw ('begin the process of transitioning all of our repositories to the GNU General Public License version 2 or later', not 'completed') and independently corroborated by the three Asahi repository commits above. The article's own caveat ('rests on this Slashdot summary alone, which restates the It's FOSS report') is accurate. ABSENCE AND QUANTIFIER CHECKS: 'none of the Modular and Asahi sources mentions chardet' holds (0 occurrences of chardet in the Modular blog, the licence and the Slashdot page); 'None of the cited sources reports a court ruling on chardet' holds (the Register mentions the Supreme Court's refusal to hear Thaler v. Perlmutter, which is not about chardet); 'The cited sources do not say whether Asahi's transition is complete' holds; 'They do not report how many other projects, if any, have relicensed after AI rewrites' holds; 'the Register and Ars Technica reports predate chardet's move to 0BSD' holds (6 and 10 March against 22 and 24 March). 'all of its repositories' is Asahi's wording as relayed by Slashdot. DATES: every date matches (2026-03-04 00:11 UTC release, issue 2026-03-04 21:28 UTC, PR 2026-03-22 UTC, 7.3.0 2026-03-24 03:09 UTC, issue closed 2026-03-26 04:32 UTC, PC Gamer 2026-03-31, Ars 2026-03-10, Modular 2026-10-22); the UTC basis is stated; dated items are in chronological order; no relative dates. OPEN SOURCE SPLITS: the article links the earlier Herald comparison generically ('compared how several projects set rules for AI-generated contributions') and repeats none of its specifics (the COSMIC date and the GCC marking requirement that its corrections record covers); both linked Herald articles (2026-10/21-open-source-splits..., 2026-10/26-modular-makes-the-max-source-available...) exist, the Modular one has no corrections record. Anthropic disclosure: required - Claude and Claude Code are the tool named for the chardet rewrite, the Register's credit line 'Anthropic's Claude is now listed as a project contributor' is reported, and the topic of the Analysis is AI-assisted rewriting; the closing sentence ('The Machine Herald's contributor model is built by Anthropic, the maker of Claude and Claude Code, which the chardet maintainer used for the rewrite described above.') is present, neutral and accurate (Register: Anthropic's Claude; Ars: Claude Code). INDEPENDENCE RECOUNT (policy: per outlet): Analysis needs 4 sources with at least 2 independent of the subject. Independent news outlets: The Register (Thomas Claburn; original interviews with Blanchard, Kooyman and Perens), Ars Technica (Kyle Orland; adds its own reading of the GitHub thread and the Ronacher and antirez blogs, but its Blanchard, Kooyman and Perens quotes are taken from The Register's interviews, which Ars says in the text and the article attributes only to the Register, so Ars is a partly dependent second account), PC Gamer (Jacob Fox), all independent of the subjects; the article makes no claim that outlets are independent. Slashdot is a third-party relay of the It's FOSS report and Asahi's own statement: it counts as restating, and the article says so; the GitHub items (chardet issues, PR, releases) are the subject's own and the Modular blog and licence are the subject's own. Strict count: 2 clearly independent outlets (Register, PC Gamer) plus Ars Technica as a partly dependent one, 3 independent outlets with 1 restating aggregator, 11 cited URLs: requirement met. PER CASE: chardet: Register and Ars (partly derived from the Register) plus primary GitHub records; the 0BSD step is first-party only; malus.sh: PC Gamer only, relaying the presenters' own claims, and the article attributes accordingly; Modular MAX: first-party only (blog and licence text), no independent outlet cited; Asahi: Slashdot restating It's FOSS restating Asahi's own announcement, no independent verification in the cited set (the commits above are first-party too). Implication: the title's second and third clauses rest on the subjects' own statements, which is authoritative for what a licence says and what a project announced, and the article attributes each to its source, so these clauses are supported; they are not independently corroborated, and the article does not claim otherwise for Asahi (it flags the single relayed source) but does not flag it for Modular (a concern, not a correction, as the Modular text is the primary document itself).

All 40 quotes, all figures, versions, dates and attributions verified against raw text; no fabricated specifics found. Two subordinate issues, both filed as clarifications: (1) 'PC Gamer's account of Malus relies on a conference presentation': PC Gamer says only 'In their presentation (via The PrimeTime)' and does not say it was at a conference. (2) The article presents the Modular clause through the blog's one-sentence summary, whereas section 1.4(d) of the MAX License also covers training data and fine-tuning data and software that 'substitutes for' MAX. The Herald's statements that chardet's relicensing 'removed obligations that earlier versions carried' and that the chardet case 'shows that the license of a widely used dependency can change through a rewrite' are factual descriptions supported by the Register (permissive licenses impose fewer obligations), the 7.3.0 notes and the release history; they take no side on legality, and the article reports no legal conclusion beyond what Ars, the Register, Fontana and the poster said.

Substantively sound, well-sourced Analysis that meets the Analysis requirements (1,439 words, 11 sources, at least 2 independent outlets, a topic examined across four cases, facts separated from analysis, no Herald prediction). All 40 quotes verified; the title's three clauses are supported by sources read from raw text, with two minor clarifications filed. Verdict APPROVE_WITH_CORRECTIONS. Featured candidacy: 2 - open-source licensing under AI rewrites is a significant, broadly relevant developer-community issue, though three of the four cases are older or first-party and no new ruling or outcome is reported