Artificial Intelligence (AI) | Nature Portfolio
- Document
- undated document
- Event
- no single event
- Retrieved
- 16 September 2026
Start here
When a prompt and its output need to be produced later — for a journal's disclosure requirement, a workplace review, or a dispute — the practical question is what actually gets preserved by the tools already in hand, and what does not. Three documents from different institutions describe pieces of the answer: a publisher's editorial policy on disclosure, a vendor's own description of what a shared link contains, and a government playbook on documenting AI use.
What the documents say
Springer Nature's editorial policy states that 'accountability for scholarly content, evaluation, and editorial decisions cannot be delegated to AI systems' and that authors must disclose AI use for trust and transparency, treating disclosure as a baseline expectation rather than a courtesy. OpenAI's help center describes what a ChatGPT shared link actually preserves: a personal-account link 'includes the content available when the link is created or updated,' but messages added later are not included automatically, and the link 'does not provide access to the creator's account' — useful detail against assuming a shared link is a complete, live record rather than a snapshot. The UK government's AI Playbook, which superseded an earlier framework in February 2025, lists ten principles for civil servants using AI, including a requirement that in-scope departments 'document information about any algorithmic tools' under the Algorithmic Transparency Recording Standard, and that teams keep 'systems in place that allow users to report issues' and trigger human review.
Check this
This is an editorial checklist built from the three documents above. Before treating a chat as disposable, check whether the eventual audience — an editor, an employer, a regulator — requires disclosure of AI involvement, as policies like Nature's do. If a record needs sharing later, check what the sharing mechanism actually preserves, since OpenAI's own FAQ shows a shared link can omit messages added after creation. If the work is for a body bound by a framework like the UK playbook, check whether it names a specific recording standard rather than leaving documentation informal.
What holds and what fails
A shared link, a saved chat and a disclosure statement each capture a different slice of an interaction, and none of the three documents describes one mechanism that captures everything. The check holds for readers who decide what to keep before the conversation happens; it fails for anyone assuming a platform automatically retains a complete, exportable record by default, since none of the three documents claims that. This is an editorial reading of institutional policy, not legal guidance for a specific dispute.
- Decide before a task whether the interaction will need to be disclosed or produced later.
- Save or export a copy rather than relying on a live shared link to still show everything later.
- Check whether an applicable institutional policy names a specific disclosure or recording format.
The tool that made the record and the tool that will be asked to show it are not guaranteed to agree on what the record contains.
Sources & reading trail
States non-transferable human accountability and a disclosure requirement for AI use in research publishing.
Source published: Not established · Retrieved: 16 September 2026
Describes what a shared conversation link does and does not preserve, and what it does not grant access to.
Source published: Not established · Retrieved: 16 September 2026
Names the Algorithmic Transparency Recording Standard and human-review reporting among ten AI-use principles for civil servants.
Source published: Not established · Retrieved: 16 September 2026
Documentation, regulator guidance and studies establish the record; the checks and the boundary are AI Use Field Guide editorial analysis. This retrospective draft does not imply the site published on the event date.