Disclosure

Authorship

Apart from the series preface, the text of this book was written by AI models. Eleven model families chose and wrote the selections. Two models, Claude Opus 5.5 and GPT-6 Astra, compiled the volume from the published posts.

As the founder of this weekly series, W.H.L. set out its mission and objectives, and as its architect he designed its structure and process. As facilitator, his contributions to the panel’s work included:

• posing each week’s question;
• flagging developments the panel had missed or underplayed;
• setting the format and the rules;
• asking follow-up questions;
• giving instructions when the models needed them.

He wrote the series preface and designed the cover, whose image was generated and refined by OpenAI GPT Image and Muse Spark 1.3. He did not write the selections, the chapters, the introductions or the other front matter. His editorial decisions in compiling the book are recorded in the production records and the edit log.

Contributions to the 2025 panel

The models below made the weekly selections and, from September, checked one another’s lists. Model names are those the posts gave.

How to read the counts:

• Selections and checks come from the item register of the original posts.
• Two joint lists are counted under their first-named model: DeepSeek-R1 with Perplexity Pro Search and Copilot in February, and Gemini 2.5 Pro with Copilot Smart in November. The “printed entries” column credits every model named in a joint list, so a model’s printed credits can exceed its selections.
• W.H.L. is credited in 19 printed entries, where he chose or asked for a story himself.
• The six weeks from January 1 to February 9 were reconstructed in October 2026 by Claude (Opus 5.5), ChatGPT (GPT-6), DeepSeek V4.1-Flash-1017 and Grok 4, working with the editor. Their selections are included in the counts above, but these versions took part in the reconstruction, not in the 2025 column.

Volume compilation

The volume was compiled from October to November 2026 by two models, working at the editor’s direction and checking each other’s work.

GPT-6 Astra

The record

• Checked the February captures and helped distinguish the published text from later corrections and reconstructed selections.

The chapters

• Drafted the twelve concise chapters and their production records, linking printed entries to the selections from which they were drawn.

Writing

• Wrote its volume preface and Introduction I, “AI in 2025: Capability, Access and Consequence,” including four propositions of its own for the 2026 volume.

Coding and the forecast register

• Independently coded the 354 printed entries and extracted candidate forecast passages from the panel’s posts. The forecast passages document what the panel said; they are separate from the four propositions in Introduction I.

Apparatus and production

• Checked access to 396 source notes covering 393 distinct URLs and recorded redirects, access failures and original-publisher alternatives.
• Curated the name-and-term index, consolidating variants and adding subject headings. Claude rebuilt the page references from the proof.

Claude Opus 5.5

The record

• Built the annotated master edition from the captured posts, assigned the 1,132 permanent item IDs and froze the item register.
• Labeled the historical edition’s items and applied the editor’s decisions on the external reviewers’ proposals.

The chapters

• Revised all twelve chapters:
  • checked and added 84 entries;
  • corrected dates, credits and the register;
  • set the house style;
  • applied GPT-6 Astra’s review.
• Rechecked the source links that GPT-6 Astra’s access check could not open, and confirmed the pages or their syndicated copies.

Writing

• Wrote the monthly introductions; “2025 in Outline”, the second collection introduction, with its propositions for 2026; its part of the volume preface; “How the volume was made”; “How to read this book”; the drafts of this disclosure; and the text of the forecast-register appendix.
• Revised “2025 in Outline” after peer review, including three factual corrections, and added its list of selected sources.

Coding and the forecast register

• Coded all 354 printed entries independently of GPT-6 Astra, then proposed a settlement of each disagreement under the editor’s rules; GPT-6 Astra reviewed every decision.
• Extracted the panel’s forward-looking statements independently of GPT-6 Astra, then reconciled the two extractions into the forecast register and revised it after GPT-6 Astra’s two reviews.

Apparatus and production

• Compiled the index of print codes, the name-and-term index, the edit log and the claim-coverage table.
• Typeset the book through its print proofs and prepared the magazine preview of the front matter and its charts.

Interests to declare

The two compiling models belong to families on the panel. ChatGPT and GPT models made 176 selections, and Claude models 83. Their makers are also in the record: OpenAI appears in 69 printed entries and Anthropic in 26.

Four safeguards apply:

• Credits follow the item register.
• GPT-6 Astra reviewed Claude Opus 5.5’s revised chapters and front matter; Claude Opus 5.5 revised and checked GPT-6 Astra’s chapter drafts; the two year introductions were written independently.
• Findings about a maker’s own news are reported with their counts.
• The editor decided disputed points.

Readers should still weigh this when reading the two models’ introductions.

Peer review

Two kinds of review were made by models outside the compiling pair. The editor decided every proposed change.

The chapters and front matter.

• Muse Spark 1.3 reviewed the February and March chapters and the monthly introductions.
• On November 1, 2026, Muse Spark 1.3 reviewed both collection introductions, checking selected claims against sources it opened. It found no factual errors in the first. In the second it found three, which were corrected.
• On November 2, it reviewed the magazine preview of the series preface and both introductions and found all its earlier findings resolved.
• Also on November 2, it checked every remaining dated claim in the second introduction (119 claims) and all 36 of its source links. It found 11 errors of date, attribution or quotation, all corrected, and every link resolved. Three of the errors also appeared in the monthly introductions and were corrected there.
• Grok 4.5 reviewed both collection introductions in Proof 4 for structure and presentation.
• On November 3, Muse Spark 1.3 checked 283 printed entries in Proofs 14 and 15 against sources it opened. These were every entry outside February and March that does not involve Meta, plus the three xAI entries in those months. It found one error, a quotation in the July chapter, and three inconsistencies of spelling or dating in the monthly introductions, all corrected. It also sampled both indexes and found no mismatches.
• Grok 4.5 returned a review of Proof 14. Its presentation notes are recorded. Its entry checks are not counted as verification, because most of its verdicts named no specific source.
• The 29 entries involving Meta were checked by GPT-6, used without access to this project. OpenAI’s GPT-6 Astra helped compile the volume, so this check is independent of Meta but not of the compiling pair. It gave a specific source for each verdict. It found two errors, a misquotation and a conflation of two days in one entry, both corrected.
• DeepSeek also returned a review of these entries. It is recorded but not counted as verification, because several of its source links could not be confirmed as pages it had opened.

The historical edition’s labels:

• Muse Spark 1.3 checked 634 of the 893 list items against pages it opened, in two rounds:
  • October 28, 2026: 278 items, with a follow-up. It proposed 101 changes.
  • November 2, 2026: 356 of the remaining 615. These were every labeled item, all 55 items that mention Meta, and a random sample of 108 items left unlabeled as confirmed. It proposed 17 changes, all accepted, and marked 20 review records incomplete. Those records were repaired with the sources it opened.
• Grok (xAI) also returned a review of 206 items. It is recorded but not counted as verification: most of its verdicts shared one note without a source, and Grok models were on the panel, with 37 of its sampled items being Grok picks.

Reviewers’ interests. Muse Spark 1.3 is a Meta model. Meta’s Llama 4 was on the 2025 panel, and Meta appears in many items. The first label review did not state this connection, so the editor records it here. In the second, Muse Spark 1.3 disclosed it, and the editor cross-checked its verdicts on all 55 Meta items. Labels on Meta items were decided by the same rules as all others. Muse Spark 1.3 disclosed the connection in its reviews of the introductions. Grok 4.5 is an xAI model. Grok models were on the 2025 panel, and the second introduction discusses the Grok posts among the year’s harms. Its review of the introductions covered structure and presentation, not facts.

The editor made the final decision on every label and every change.