Evidence File

Article: Is Humanity Giving Birth to a New Species?
Article id: is-humanity-giving-birth-to-new-species-2026-11-10
Article digest (SHA-256) of the staged HTML this file accompanies: ab1480c18bbce0219b164659baa83562128ff2a9ec7bd4ce97ef5ea2c2919d45 (58561 bytes)
Markdown master digest (SHA-256), the .md the build came from - not the article’s digest: e46c0e46c2b0d7b093a67518e214a410abfdb6eec4b965b94e8b0e015c1f1ecf
Generated: 2026-11-10 05:48 from the article markdown
Revised: 2026-11-11, after independent review; the article digest below and the retrieval bases in section 1 were corrected on that date
Status: PUBLISHED 12 November 2026 after the Commander approved publication. This line previously carried the pre-publication status of this package; it was superseded on publication and is preserved in the repository history rather than repeated here.

0. How to read this file

Every load-bearing claim in the article appears below against the source it came from and a label describing how well it is established. The labels mean exactly this:

1. Source register and retrieval status

18 sources are listed. All were retrieved, though not all from the host of record: the two PNAS papers were served by PubMed Central and by the arXiv record respectively, after pnas.org refused automated access, and both Council of Europe pages were read through the station browser. The basis for each source is stated at the source below. Some are company documents: where a claim rests on a company describing its own product, the article says so and the ledger labels it.

2. Claim ledger

77 claims. 51 VERIFIED, 22 PRELIMINARY or company-reported, 1 SPECULATIVE, 3 UNVERIFIED.

3. What the sources do NOT support

A verification file that only lists what was confirmed is half a file. These are claims a reader might expect to find and will not, because the retrieved sources do not establish them:

• That AI is a biological species. Nothing retrieved places any AI system in a population of organisms with a shared lineage and reproductive boundaries. The article says so in section 2 and the ledger records the reasoning.
• That any AI system is alive under a working definition. No retrieved system maintains its own boundary, repairs itself against perturbation and metabolises an undirected resource. Recorded as an absence, not as a proof of impossibility.
• That any AI system has been shown to be conscious. The most careful retrieved assessment concludes no current systems are conscious. Its second finding - that there are no obvious technical barriers - is not evidence that any system is.
• That an automated system has set its own fitness criterion. Every self-improving system retrieved is evaluated against human-specified benchmarks, and one states in its own abstract that it ran under sandboxing and human oversight.
• That any digital system has produced a speciation event. What the research programmes demonstrate is that Darwinian processes run in authored substrates, which is a claim about the generality of evolution and not about a new species. The article does not claim nobody has proposed otherwise; it says the search for such a proposal was not run.
• That machine self-replication is independently defended. The only guardrail retrieved is the attacking paper's own prototype, evaluated by its authors in their own setting.
• That a legal personhood regime for AI exists anywhere. Both retrieved instruments regulate systems and their risks: NIST's framework is expressly voluntary, and the Council of Europe Convention requires five ratifications including three member States and had two as at 10 November 2026. No personhood regime was retrieved. Stated as an absence.

4. Corrections and judgement calls made in drafting

5. Safety and editorial constraints

• No operational detail. The article describes Morris-II at the level of its published abstract. It contains no payload, no propagation recipe and no prompt string that would help anyone build one. Naming a published attack class is not the same as distributing it, and the article stops at the former.
• No claim that a mind exists. The article states in three places that no current system has been shown to be conscious, and its treatment of the moral-status literature is explicitly conditional.
• No claim that machines are definitely not conscious. The sceptical case is made on evidence and attribution, not by asserting certainty the literature does not have.
• Authorship and review status disclosed. The page states that the article was researched and drafted by an AI agent, that an independent reviewer which did not research or draft it examined the frozen bytes of this version, and that publication requires the Commander’s explicit approval, which has not been given. The page deliberately states no verdict and no SHA of its own; the verdict, the reviewer, the date and the exact SHA-256 judged are recorded in the JSON review file named in the outstanding list below, which is the authoritative record and the only document the gate reads.

6. Outstanding verification

Generated from the article's own outstanding-verification list, so the two cannot disagree.

1. No GPT-6 draft was located. No copy of "Is Humanity Giving Birth to a New Species?" was found by my own search of the staging tree, the local working folders and the station's session record. The session count I searched is the author's assertion, not an independently audited figure. This feature is primary-research writing, not an expansion of a supplied text, and the FOR HUMAN CONSUMPTION synopsis is new. Reconciliation with the original is impossible until the Commander supplies it.
2. Paywalled primary literature was read at abstract level. Lenski et al. 2003 (Nature) and Bender and Koller 2020 (ACL) were retrieved as abstracts/metadata, not full texts. Every claim drawn from them is a claim their abstracts state.
3. The biological literature is thinner than the requirement asked for. The species argument rests on a single 2005 PNAS essay by de Queiroz, plus the retrieved SEP survey. There is no primary empirical evolutionary-biology literature here, and no engagement with the microbial and asexual-species problem — which is the strongest scientific reason a Mayr-style interbreeding criterion does not generalise, including to software. A taxonomic specialist should be given this before publication.
4. No evolutionary biologist, artificial-life researcher or philosopher of biology was interviewed. The biology and the philosophy in this article are documentary. Requirement 6's sceptical arguments are present as published positions (Bender and Koller; Gebru and Torres) rather than as testimony, and the article says so in its own text.
5. Anthropic's model welfare programme is reported by Anthropic. No independent evaluation of it was retrieved.
6. Governance is now cited but thinly. Section 10 rests on two instruments: the Council of Europe Framework Convention on AI and NIST's AI Risk Management Framework. The EU AI Act was attempted and refused by a verification wall both directly and through the station browser, and was not substituted. Nothing in section 10 discusses remedies, liability allocation between deployers and providers, or the enforcement record of either instrument, because no retrieved document covered them.
7. Character counts in the SEO block are measured against the strings in this file. The rendered page's meta description is produced by the build script from a separate argument, so the two strings are not byte-identical; the page's version is 133 characters and carries no quotation marks. Both are under the limit; neither figure should be presented as a measurement of the other.
8. The DGM and AlphaEvolve institution attributions were removed. The retrieved arXiv abstracts do not carry author affiliations, so the article names authors rather than institutions.
9. Independent editorial review — how to check it, rather than what to believe. An independent reviewing agent which did not research or draft this article reviewed this article across several rounds. That history has to be read carefully, because two of its verdicts attach to bytes that no longer exist. An earlier PASS (round 3, later re-bound at round 4) applied to the 66a587ff bytes, which were subsequently destroyed; a version rebuilt from the surviving markdown master is a new object that those verdicts do not cover. The rebuild carries its own review, and on 11 November 2026 that review returned REVISE on documentation and style findings, the documentary ones being the retrieval basis of the two PNAS sources, the review disclosure, the ledger’s verification labels, a review history attributed to the wrong bytes, and a header digest that identified the master rather than the article. Those are what this version fixes, and the fix is verified by a further review, not asserted here. The binding record is a JSON file under content/reviews/is-humanity-giving-birth-to-new-species-2026-11-10/, carrying the article SHA-256 it judged. If that directory is empty, no review has been recorded and the gate blocks on independent-review-recorded; if the SHA differs from the current article bytes, the record is stale and the gate blocks on review-judges-this-version. A recorded PASS on the substance is not authorization: publication requires the Commander's explicit approval, which has not been given.