Skip to main content
v2026.11,986 entries · CC-BY 4.0
CASRAIAsk CASRAI · AI assistant for researchAsk a research question. Answers come with their sources.Funder rules, compliance, methods and publishing. Answers link the sources they rest on.
NIKOLAI elementN8 · Transparency and reviewProposednikolai-v0.3

Redaction

Definition

A redaction is the removal of information from the public version of an AI safety document, such as a risk report or model card, recorded with where it occurred, how much was removed, why, who decided, and a notice that it happened.

Also called: Redaction record, Redaction disclosure, Redaction summary, Redaction disclaimer

NIKOLAI's proposed operational definition

NIKOLAI editorial proposal (unsourced -- NIKOLAI's own reading, not any org's declared position): a redaction record is a recorded removal of information from a published artefact, capturing where the removal occurred (location within the document), how much was removed (extent), why it was removed (reason, drawn from a controlled redaction-reason vocabulary), who decided to remove it (deciding party), and a disclosure that the removal occurred at all -- distinct from silently omitting the information.

The operational definition is CASRAI's own proposal, not a definition any named organisation has agreed to. See what NIKOLAI is and is not, or browse the AI safety glossary.

Examples

Examples from published documents

  • A frontier developer that redacts a published document must describe the character and justification of the redaction and retain the unredacted information for five years (California SB 53, 22757.12(f)(2)).
  • Anthropic will disclose the existence of each redaction in the public Risk Report and aim to give a brief justification (Anthropic RSP v3.4, §3.5).
  • METR's incident report opens with: 'Except where explicitly noted in this post, OpenAI redacted no additional information that was important to our conclusions' (METR, OpenAI-Hugging Face incident investigation).

Counter-examples

Not to be confused with

  • Silent omission, where information is simply left out with no notice, is not a recorded redaction; the element exists to make removals visible.
  • Summarising a result instead of reporting it in full is related but distinct; the OpenAI Preparedness Framework allows disclosures to be 'redacted or summarized'.

Related terms

Related AI safety terms

Terms defined under this element that are not NIKOLAI elements themselves, each as the cited document uses it. Every term is also listed in the AI safety glossary.

Redaction summary statement
A redaction summary statement is a short notice at the top of an evaluator's report saying whether the developer's redactions removed anything important to the evaluator's conclusions.
Source: METR: OpenAI-Hugging Face incident investigation
Redaction disclaimer
A redaction disclaimer is a notice on an evaluator's publication stating what authority the developer or another party had to edit or redact it, whether that authority was used, and what was removed.
Source: AEF-1: Minimum Operating Conditions for Independent Third Party AI Evaluations (v1, 4 Dec 2025)
Unredacted copy
An unredacted copy is the complete version of a safety document from which a public version was redacted; California SB 53 requires a frontier developer that redacts a published document to retain the unredacted information for five years.
Source: California SB 53

Source of record

Where this definition comes from

Crosswalk

How named organisations use this concept

Every row below is a shadow mapping. A shadow row is CASRAI's own reading of a published document. No lab, evaluator or regulator named on a shadow row has declared, endorsed, or been consulted on it. That changes only when an organisation files its own Mapping Declaration.
OrganisationTheir term, as publishedMatch & verificationSource
AnthropicShadow mapping
Anthropic RSP v3.4 / Risk Report Aug 2026 / Advanced AI Framework / We Must Pace the Frontier (essay by Anthropic's CEO)
“"Version 3.4 of our RSP requires us to publicly disclose, at a high level, when we make redactions in the public version of a risk report" (§1.3.4); "We will disclose the existence of each redaction made in the public version of the report and aim to give a brief justification for such redactions, though in many cases this will necessarily be very high-level" (§3.5). AAF: "Public-facing materials should be redacted only as necessary, and developers should note the nature of the redaction in any published version." (p.7). Essay: "We will have the narrow ability to redact security-sensitive, legally privileged, commercially sensitive, or third-party confidential information".”
Row also draws on the Anthropic Risk Report Aug 2026 (the §1.3.4 sentence is from the Risk Report), the Advanced AI Framework, and the 'We Must Pace the Frontier' essay; citation pinned to RSP v3.4 as the most specific verbatim commitment.
exactEQ
confidence: high
Anthropic Responsible Scaling Policy v3.4
OpenAIShadow mapping
OpenAI Preparedness Framework v2
“"Such disclosures about results and safeguards may be redacted or summarized where necessary, such as to protect intellectual property or safety." (§5.2)”closeCL
confidence: high
OpenAI Preparedness Framework v2
xAIShadow mapping
xAI Risk Management Framework (20 Aug 2025) / Frontier AI Framework (30 Jun 2026)
“Aug 2025 RMF: transparency items xAI "may publish", with redaction, and unredacted versions to vetted red teams or government agencies. June 2026 FAIF: no redaction clause found.”
Also cites xAI's 30 Jun 2026 Frontier AI Framework, whose PDF metadata /Title reads 'Privileged/Confidential DRAFT working FRAMEWORK DOC' with no xAI statement found disambiguating draft vs final status. The finding that the June 2026 document contains no redaction clause is keyed to that unresolved-status document and should be treated as provisional.
closeCL
confidence: medium
xAI Risk Management Framework, 20 Aug 2025
MetaShadow mapping
Meta Advanced AI Scaling Framework v2
“"Preparedness reports will provide as much detail as possible, while redacting information to protect trade secrets or as appropriate under law. If we redact information, we will describe our reasons why to the extent that we can without creating undue risk." (§2.2.1)”exactEQ
confidence: high
Meta Advanced AI Scaling Framework v2
California SB 53Shadow mapping
California SB 53
“"If a frontier developer redacts information in a document pursuant to this subdivision, the frontier developer shall describe the character and justification of the redaction in any published version of the document to the extent permitted by the concerns that justify redaction and shall retain the unredacted information for five years." (22757.12(f)(2))”exactEQ
confidence: high
California SB 53
EUShadow mapping
EU GPAI Code of Practice, Safety and Security Chapter
“Framework and Model Report notifications to the AI Office are given "without redactions, unless they are required by national security laws to which Signatories are subject" (Measures 1.4, 7.7); serious-incident information is reported "redacted to the extent necessary to comply with other Union law applicable to such information" (Measure 9.2). No public-version redaction-disclosure rule equivalent to Anthropic's or SB 53's was found in the chapter; disclosure to the AI Office is expected to be complete.”closeCL
confidence: high
EU GPAI Code of Practice, Safety and Security Chapter
METRShadow mapping
METR OpenAI-Hugging Face incident investigation report
“"Redaction summary statement": "Except where explicitly noted in this post, OpenAI redacted no additional information that was important to our conclusions."”exactEQ
confidence: high
METR: OpenAI/Hugging Face incident investigation
AI Evaluator Forum (AEF-1)Shadow mapping
AEF-1: Minimum Operating Conditions for Independent Third Party AI Evaluations
“AEF-1 principle 4 conditions "Redactions" and "Redaction disclaimer": the evaluator discloses "the redaction authorities granted to the system provider or other external entities".”
Verified against the AEF-1 PDF (v1, 4 Dec 2025).
closeCL
confidence: high
AEF-1: Minimum Operating Conditions for Independent Third Party AI Evaluations (v1, 4 Dec 2025)
AI Evaluator Forum (AEF-1)Shadow mapping
AEF-1: Minimum Operating Conditions for Independent Third Party AI Evaluations
“"4.6 Redactions Requirement: The system provider did not have authority to redact results to conceal concerning findings"; "4.7 Redaction disclaimer Recommendation: The evaluator clearly disclosed the redaction authorities granted to the system provider or other external entities", stating "whether the authorities were exercised or not, as well as describing what was redacted".”
Evaluator-side standard: governs redactions a developer makes to an evaluator's publication, not to the developer's own report.
exactEQ
confidence: high
AEF-1: Minimum Operating Conditions for Independent Third Party AI Evaluations (v1, 4 Dec 2025)
What do these codes mean?
exact
The source term is equivalent to this element
close
The source term is close but not equivalent to this element
broad
The source term is broader than this element
narrow
The source term is narrower than this element
none
No mapping claim — used for false-friend and declared-but-undefined rows
EQ
Equivalent
CL
Close
BR
Source is broader than the element
NR
Source is narrower than the element
FF
False friend — same or similar label, different meaning
DU
Declared but undefined by the source
UV
Unverified

Gap

Redaction rules exist in Anthropic's RSP, Meta's framework, OpenAI's Preparedness Framework, SB 53 and the EU Code of Practice, but none specifies a structured record of location, extent, reason code and decision-maker. The EU Code of Practice expects unredacted disclosure to the AI Office rather than a public redaction notice.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →