Redaction reason

NIKOLAI editorial proposal (unsourced -- NIKOLAI's own reading, not any org's declared position): a redaction reason is a controlled value naming why a piece of information was redacted from a published artefact. The vocabulary should carry both permitted values (e.g. legal compliance, intellectual-property protection, public-safety considerations, privacy, cybersecurity, national security) and, distinctly, prohibited values -- reasons a developer may not cite, such as redacting a finding merely because it is unfavorable.

This is CASRAI's own proposed definition, not a definition any named organisation has agreed to. See what NIKOLAI is and is not.

Source of record

Where this definition comes from

Anthropic Responsible Scaling Policy v3.4, §3.5
“1. Legal compliance: To comply with legal obligations on disclosing information, such as export control regulations, national security restrictions, or contractual obligations with third parties. 2. Intellectual property protection: To protect proprietary information, including technical details of our models and training methodologies, trade secrets, or other confidential business information. 3. Public safety considerations: To protect public safety by not disclosing information that could be exploited to cause harm. 4. Privacy: To protect the personal privacy of individuals.”
https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf
California SB 53, 22757.12(f)(1)
“necessary to protect the frontier developer's trade secrets, the frontier developer's cybersecurity, public safety, or the national security of the United States or to comply with any federal or state law”
https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53
OpenAI Preparedness Framework v2, §5.2
“such as to protect intellectual property or safety”
https://cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf

Crosswalk

How named organisations use this concept

Every row below is a shadow mapping. It is CASRAI's own reading of a published document. No lab, evaluator or regulator named here has declared, endorsed, or been consulted on this mapping. That will change only when an organisation files its own Mapping Declaration — see the non-endorsement policy.

Organisation	Their term, as published	Match	Source
Anthropic Anthropic RSP v3.4 / Pace the Frontier essay / Advanced AI Framework	“RSP v3.4 §3.5 four verbatim reasons: "1. Legal compliance... 2. Intellectual property protection... 3. Public safety considerations... 4. Privacy..." Amodei's permitted grounds: security-sensitive, legally privileged, commercially sensitive, third-party confidential; prohibited: "we can't redact findings just because they are unfavorable". AAF: "trade secrets", information "that would materially compromise public safety, model security, or national security", or to "comply with applicable law" (p.7).” Also draws on {PACE} (Amodei's permitted/prohibited grounds) and {AAF} (p.7); citation pinned to RSP34 as the verbatim four-item list, which also uniquely names a prohibited reason (see gap_note).	exact confidence: high	Anthropic Responsible Scaling Policy v3.4
OpenAI OpenAI Preparedness Framework v2 / GPT-5.6 Safeguards Report	“"such as to protect intellectual property or safety" (PF §5.2); Safeguards Report details "not suitable for public disclosure (such as information potentially useful to attackers)" (GPT-5.6 s.9.4).” Also cites {G56}, the GPT-5.6 deployment safety / model card page, for the s.9.4 detail.	narrow confidence: high	OpenAI Preparedness Framework v2
Meta Meta Advanced AI Scaling Framework v2	“"to protect trade secrets or as appropriate under law" (§2.2.1).”	narrow confidence: high	Meta Advanced AI Scaling Framework v2
California SB 53 California SB 53	“a frontier developer may redact information "necessary to protect the frontier developer's trade secrets, the frontier developer's cybersecurity, public safety, or the national security of the United States or to comply with any federal or state law" (22757.12(f)(1))” Exact statute text confirmed directly from leginfo.legislature.ca.gov in the source research's fourth verification pass (16 Sep 2026).	exact confidence: high	California SB 53
EU EU GPAI Code of Practice, Safety and Security Chapter	“Narrower than the lab/statute lists: reporting to the AI Office is expected "without redactions", with only two named exceptions -- "national security laws" (Framework/Model Report notification, Measures 1.4, 7.7) and what is "necessary to comply with other Union law applicable to such information" (serious-incident information, Measure 9.2). Confidentiality (including commercial confidentiality) is separately "preserved to the extent required by Union law" for information the AI Office holds (Recital (d)).”	close confidence: high	EU GPAI Code of Practice, Safety and Security Chapter
METR METR OpenAI-Hugging Face incident investigation report	“OpenAI redactions in the incident report were "for IP" (para.).” Paraphrase ('para.') rather than an exact quoted sentence in the source document.	close confidence: medium	METR: OpenAI/Hugging Face incident investigation