Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products
Document status what the publisher asserts
draft — recorded as draft against the primary source
No status read is recorded for this entry — this record has not read a publisher surface for it, which is a fact about this record and not about the publisher. A status read is recorded for 7 of 132 tracked entries, 5 of which banked a passage.
Evidence state what this record holds
verified by a person — a person checked this entry against the primary source on a recorded day
/api/v1/instruments.json.
What this record holds for this entry, rung by rung, each mark derived from a field, a count or a stored reading. Rungs in the order the record acquires them, from what an entry is to what has been checked about it. The order is fixed; a rung’s position says nothing about any entry.
✓ held — this record holds it, and the sentence beside it names what from · ✗ read, and not there — a reading ran and did not find it — an absence this record measured · — nothing held — no reading of this rung is recorded — nobody has looked, which is not the same as looking and finding nothing
The first rung a reading ran on and did not find: type — The title was read and it does not call this document a regulation, a guidance or a standard, so no class is recorded.
Each term states what this record does for this entry and what it does not, derived from the entry’s own fields and its own series. It is a record of coverage, not a finding about the document.
The test an entry is admitted under is published in full, with its negative limb and its worked examples, at the inclusion rule. What this record holds about THIS entry's admission is below, each line naming the store it was read from.
Filed core. For that label this archive undertakes to put an entry's status to a person. The label records what this archive undertakes for a class of entries. It states no finding about this one, and it is not a ranking.
Resolution: listing resolved · document resolved to the retrieved document (opens in new tab) fda.gov
What this archive can and cannot establish about this entry. What this row hashes is FDA's guidance search index, not the guidance. All 117 observation records name that index — https://www.fda.gov/regulatory-information/search-fda-guidance-documents — as the url fetched, and 116 of the 117 carry one unchanging digest, which is that index page's. The guidance's own page is recorded as `resolved_url` and is not what the series hashes. The two text changes this page reports are one excursion and its return, not a change to the guidance: on 2026-08-20 the request was answered at https://www.fda.gov/about-fda/page-not-found under HTTP 200 with a 722-character body and a different digest, and on 2026-08-21 the series returned to the index digest. The status does not rest on these bytes — it was confirmed by a person on 2026-07-30 from FDA's own field `field_final_guidance_1` in the guidance datatable row for this document, at the guidance's own URL. Whether the observation target should move to that URL is a resolution-target question; nothing here settles it. Account written 2026-08-25; this row carried none before.
Recorded as draft (draft) against the primary source on 30 July 2026.
The supporting text is not reproduced here: it contains terminology this site does not publish. It was retrieved 30 July 2026 by structured and is at the location below, in the primary source.
Found at: field_final_guidance_1, FDA guidance datatable row
This instrument is draft (draft). It is not settled law. The status was confirmed against the primary source on 2026-07-30; consult that source directly.
Each format carries the permalink and the date this entry was observed. The observation date is part of the citation: the underlying source may have changed since.
GxPlex. Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products [entry record]. Observed 18 September 2026. https://gxplex.com/r/fda-ai-guidance Primary source: https://www.fda.gov/regulatory-information/search-fda-guidance-documents
@misc{gxplexrfdaaiguidance,
author = {{GxPlex}},
title = {{Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products}},
year = {2026},
note = {entry record, observed 18 September 2026},
howpublished = {\url{https://gxplex.com/r/fda-ai-guidance}},
urldate = {2026-09-18},
url = {https://www.fda.gov/regulatory-information/search-fda-guidance-documents}
} GxPlex. (2026). Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products [entry record]. Retrieved 18 September 2026, from https://gxplex.com/r/fda-ai-guidance
{
"@context": "https://schema.org",
"@type": "Dataset",
"name": "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products",
"publisher": "GxPlex",
"url": "https://gxplex.com/r/fda-ai-guidance",
"temporalCoverage": "2026-09-18",
"observation": {
"url": "https://gxplex.com/observations/fda-ai-guidance#obs-2026-09-18-2",
"observedDate": "2026-09-18",
"sha256": "5c61f4c6351dd8b91debbcc4566424f1084afe7fe55e032e00c05cbc8adf7ac4",
"verifiedAt": "https://gxplex.com/verify"
},
"isBasedOn": "https://www.fda.gov/regulatory-information/search-fda-guidance-documents"
} observation.url addresses the single observation this
citation rests on.
Every observation of this document is addressed and separately citable at the observation series.
Observed on 160 occasion(s) since 2026-07-28. 160 produced a content hash. 6 change(s) to the retrieved text have been detected since observation began.
·············································································································!!········!!·····!!································
All 160 marks, one per observation, oldest first. · unchanged · ! content changed ·
~ fetched but rendered client-side, no hash ·
× could not be fetched, no hash.
Every one of the 53 days from 2026-07-28 to 2026-09-18 holds an observation.
2 of these 160 record(s) are held in this series and left out of every count of the document series , each where the landing URL is a not-found path. 2 of them were hashed. They are dated to . Nothing was rewritten: each record stands in the store as it was written, with its address, its response and its digest, and the annotation sits beside it. The ruling is of . The document series this entry holds is 6 record(s), and it is what the windows in Neighbourhood below count. They are not the whole of the difference: 154 record(s) sit outside the document series, and the other 152 are classed witness, which is a reading of where a fetch landed rather than this exclusion.
Each mark links to the single observation it stands for, at
that observation’s own address. The whole series, with every address, is at /observations/fda-ai-guidance.
Any two of its retained observations can be set side by side at /compare/fda-ai-guidance.
The same evidence as a packet, with its refusals and its
integrity reference, is at /packet/fda-ai-guidance.
Every observation in this series is published, untruncated, at /api/v1/observations.json.
2026-09-18·2. /observations/fda-ai-guidance — every record of this row, with its address. /api/v1/series.jsonl — the same records for every row, one line each.
Each observation is addressed by this row’s identifier, the date it was observed, and its sequence within that date — the same key /api/v1/series.jsonl publishes as `sequence`, counted in the order the store recorded them. Counts and dates over a stated window. No figure here is a mark out of a hundred, no cell is ordered against another, and none of it is a statement about the publisher: a day this record did not reach is a fact about this record’s reach.
direct. What that record returned is in Observation history above.
https://www.fda.gov/regulatory-information/search-fda-guidance-documents 94 record(s), 2026-08-03 to 2026-09-18https://www.fda.gov/about-fda/page-not-found 2 record(s), 2026-08-20 to 2026-08-27https://www.fda.gov/media/184830/download 6 record(s), 2026-09-07 to 2026-09-18https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-use-artificial-intelligence-support-regulatory-decision-making-drug-and-biological . What sits around this entry, counted. Every figure below is drawn from a value already rendered elsewhere on this page; where a count has a list, the count is here and the list stays in the section holding its evidence.
| Window | Observations of the document | Of those, carrying a digest | Of those, carrying a change | Direct document | Official signals dated inside it |
|---|---|---|---|---|---|
| 30 calendar days | 6 | 6 | 0 | no change in window | 0 of 0 |
| 90 calendar days | 6 | 6 | 0 | no change in window | 0 of 0 |
| 365 calendar days | 6 | 6 | 0 | no change in window | 0 of 0 |
A window is marked changed where at least one observation of the document inside it recorded a change against the previous one; unchanged where the window holds at least one such observation, at least one of them carries a digest of the retrieved bytes, and none recorded a change; undigested where the window holds observations and none of them carries a digest, so no observation in it could be compared with another; and unmeasured where the window holds no observation at all. The word is shorthand for the three counts printed beside it and adds nothing to them. It is a statement about the window and never about the document outside it, which is why a window reaching back before the first observation says so underneath. A mark is derived from the observations a window holds and never from the days it spans; which days inside this entry’s own window hold none is stated by the coverage sentence in Observation history above.
The first observation of the document this entry names is . These windows reach back before it — 30 calendar days · 90 calendar days · 365 calendar days — so part of the span each names was never watched, and a window with no change inside it says nothing about the days before that first observation.
no change in window — this window holds observations of the document and none of them recorded a change — which is a statement about the window, not about the document outside it
No word is attached to the signal count beside it, and none will be. Any word for the size of a count needs either a threshold this record would have authored or a comparison against other entries that would rank the sources against one another; the count and its window are the whole claim. The document column beside it carries a word only because the rule reaching it is printed underneath and can be checked against the counts it stands for.
The last column holds no member on this entry. The zero it holds is named under “Around it” below.
topic.artificial-intelligence. Next holds the publisher-dated facts this record already carries that fall after the day this record calls today — a stated effective date, a consultation close. Nothing is projected and no step is inferred from a lifecycle: where a publisher has stated no further date, the cell says so, and what is absent is a statement rather than a future.
5c61f4c6351dd8b91debbcc4566424f1084afe7fe55e032e00c05cbc8adf7ac4 This digest is a leaf of the integrity root, and the command that recomputes it is published at /how.
The words this document uses about itself, matched verbatim against its own title. Not a statement of scope, and not a judgment about which organisations or products fall under it: each value quotes the span it was read from and links to where that span was read.
This document does not call itself a regulation, a guidance or a standard, so no class is recorded. It is filed under title states no class.
One document's retained text contained a form another document is published under, on the date shown, with the words it was read from. An edge cannot tell the manner of a citation — a document naming another in order to exclude it produces the same entry as one adopting it.
contextual mapping — not a regulatory applicability determination The policy, stated once
These citations are drawn together on the citation graph, where every row carries the same quote and source.
Listed primary first, then the further addresses this record holds for this entry, then the route it was found by, then dated third-party copies. Repeated URLs appear once, at their first position.
A field and some words a reader may keep beside this entry while reading it. This record never sees the note: nothing is sent anywhere, nothing is stored on this device, and it is gone when the page is reloaded. Nothing below is selected until a reader selects it — no default is offered here, and this record states no view about which of the four words belongs on this entry.
Those four words are a reader's own vocabulary, not this record's description of the document. A judgement of consequence, a decision to keep watching, a decision to look further and a decision to raise are refused in this record's own voice by the wall it publishes under, and that refusal is why the choice sits with a reader rather than being made here.
The copy control puts the note on this device's clipboard as Markdown,
carrying this entry's identifier and the four published addresses of its
evidence packet at /packet/fda-ai-guidance, so a note and
the packet it belongs beside can be filed together. The note is not merged
into the packet's own bytes: merging needs either a request this control
does not make, or a copy of a published surface inside every entry page,
and neither is taken.
Last updated: