Background notes
Notes the registry's chatbot reads alongside the data, the methodology and the sources: answers to questions the methodology does not cover, and the project's history. Approved by the maintainer.
Questions the methodology does not answer
Last reviewed: 15 November 2026 (John E. Milad)
Who pays for this? Is it independent? No one. It is John E. Milad's personal project, unpaid and unsponsored. It is independent of ERS Genomics, where he is CEO, and of every developer it lists. ERS Genomics licenses CRISPR/Cas9 patents; it develops no drugs and has no entry in the registry.
Can I use the data? Yes. The data are licensed CC BY 4.0: reuse them for any purpose, including commercial, with attribution. Downloads are linked at the foot of the home page: CSV files for molecules, trials, exclusions, candidates and companies, and the same data as JSON (molecules.json, trials.json) in the site's data folder.
How do I report an error or a missing molecule? Use the corrections form (aixbio-registry.org/corrections.html). Every submission is checked against its sources and the methodology, and the outcome is published in the corrections log.
How do I contact John? Through the corrections form for anything about the registry's entries; by email at aixbio.registry@outlook.com, the registry's contact address, which is also the address for privacy requests (see the privacy page); or on LinkedIn (linkedin.com/in/johnemilad).
Has John published research? Yes: six peer-reviewed papers on dialysis technology and health economics, and a granted haemodialysis patent, all listed on his ORCID record (orcid.org/0000-0002-3951-6077).
Which AI was used to build the registry? AI agents (Claude, from Anthropic) did the searching, the blind coding, the red-team challenges and the checking of quotes. In the weekly checks, a Claude model proposes a decision on each finding and TypeSafe's Jev rereads every AI claim. John made the judgement calls and signed off each strict entry. This chatbot runs on Claude Haiku 5.5.
Has anyone independent reviewed it? Not yet. Independent reviewers will be named on the About page when they have reviewed it.
How often is it updated? Each release is dated: the home page shows the data date and version. Automated checks run every Monday: trial registries, approvals, developer news, a second reading of every AI claim, and searches for new candidates. John decides each change, and approved changes are released as they are checked, often within a day. Every source was refreshed for v1.0.0 on 2 November 2026, the day the companion essay was published.
What is the companion essay? Turning the Crank Faster, John's essay on AI in drug discovery for his Substack, Backed By Reality, published 2 November 2026 alongside registry v1.0.0. It argues that AI has so far mostly sped up the existing process ("turning the crank faster") rather than shown it improves the odds, and sets out what evidence would settle it. The registry is its evidence base.
Does the registry list every AI-designed drug? No. AI input is now routine across drug discovery, and much of it is never disclosed, so no registry built from public sources can find every molecule AI touched. The registry counts a cohort: molecules dosed in people whose developers credit AI, in their own words, with choosing the target or producing the molecule. Because developers choose what to disclose, and credit AI more readily when things go well, the cohort is more likely to flatter AI than to understate it. The registry can see target novelty, phase outcomes and timing, but not cost.
Project history
Last reviewed: 15 November 2026 (John E. Milad)
- August 2026: an unpublished first count of 37 molecules, built from ClinicalTrials.gov only.
- 16 October 2026: the first full draft. It added non-US trial registries, out-licensed assets and biologics (data as of 13 October 2026).
- 19 October 2026: the domain aixbio-registry.org, and a corrections form with a private review queue.
- 20 October 2026: the draft site and its GitHub repository went public, ahead of the companion essay, so collaborators could review it.
- 20–27 October 2026: an evidence-first re-review of every entry, with two blind AI coders and an AI red team per entry, under a written review protocol that John tightened in four steps. John signed off every strict entry.
- 27 October 2026: a bibliography page of the 30 sources behind the methodology and the essay.
- 28 October 2026: this chatbot.
- 29 October 2026: the repository was made private. The registry's data, methodology, changelog and review protocol stay public on the site; the code and the project's working files do not.
- 29–31 October 2026: the registry began checking itself. A weekly, read-only watch of trial registries, news and approvals started on 29 October. On 30 October came a daily update that releases the findings John has approved; from 1 November it also releases the reader corrections he has accepted. On 31 October came a second automated reading of every entry's AI claim. Every finding waits for John's decision.
- 2 November 2026, v1.0.0: the first citable release, with every source refreshed, alongside the companion essay, Turning the Crank Faster on Backed By Reality. The drafts before it were not for citation, and the changelog starts here.
- 5–9 November 2026: a molecule-by-molecule comparison with the newest published list (Haque et al., 2026). Every molecule on it that the registry had not yet assessed went through the same review: two blind AI coders, an AI red team and John's sign-off.
- 13 November 2026: vaccines and other products whose antigens a model selected are now judged by the same rules as every other product.
- 15 November 2026: the weekly checks began looking for candidates the registry did not yet follow. They search open news, papers, and new trials by AI drug-discovery companies on ClinicalTrials.gov and, through WHO's trial portal, on other national registries.
- Next: A fuller inclusion framework (v2.0, contribution classes) is proposed but not adopted.