Safety Case (Assurance Argument)

NIKOLAI proposes to define a Safety Case as a structured, reviewable argument, composed of one or more Claim records and their supporting evidence, that a model's risks are acceptably low in a stated deployment or development context. This is a NIKOLAI editorial proposal added to the candidate element list because the sources use the label inconsistently -- some (Google DeepMind) give it a full glossary definition and instance it against a report, others (OpenAI, Meta) use an adjacent undefined term ('Safeguards case', 'safety cases') without specifying format, author, or reviewer, and Anthropic's own Risk Report does not use the label at all for what NIKOLAI would otherwise recognise as the same argument structure. It is not endorsed by, and does not restate, any single source's own definition.

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

Google DeepMind Frontier Safety Framework v3.1, glossary, p.20
“Safety Cases: are an assessable argument showing how severe risks associated with a model's CCLs have been reduced to an acceptable level. (glossary, p.20)”
https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf
Gemini 3.7 Flash FSF Report, p.41
“Our safety case relies on two evaluation suites and assumes that AI deployments will feature oversight similar to that of human employees (p.41)”
https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_fsf_report.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 Risk Report, August 2026	“Claim graph with aggregation types (§2, see C1 above). The Risk Report does not use "safety case" as its own label in the brief's record.”	close confidence: medium	Anthropic Risk Report, August 2026
OpenAI OpenAI Preparedness Framework v2	“"Safeguards case" is used in Table 2 ("require the Safeguards case to be robust to the discovered capability and/or propensity") and "not defined separately".”	none confidence: medium	OpenAI Preparedness Framework v2
Google DeepMind Google DeepMind Frontier Safety Framework v3.1 / Gemini 3.7 Flash FSF Report	“Safety case definition (glossary); "Residual risk assessments will be supplemented with a safety case when a model reaches a CCL" (glossary). The report says "Our safety case relies on two evaluation suites and assumes that AI deployments will feature oversight similar to that of human employees" (p.41) but does not define it there.”	exact confidence: high	Google DeepMind Frontier Safety Framework v3.1
Meta Meta Advanced AI Scaling Framework v2	“"Developing safety cases in advance of models exhibiting capabilities related to our enhanced evaluations." (§4.2.3); no format, author or reviewer given.”	none confidence: medium	Meta Advanced AI Scaling Framework v2
UK AI Safety Institute (AISI) UK AISI safety case templates (discovery)	“UK AISI safety case templates (inability argument) [UV].” Citation tag {DEU} does not appear in the supplied sources table and no URL could be resolved without inventing one; left blank rather than fabricated (confirmed still open in ALIGNMENT-MATRIX.md §6 item 7). Match type 'close' is inferred from context (an 'inability argument' template is directly analogous to a safety case) per the UV translation rule, not confirmed against primary text.	close confidence: low	UK AISI safety case templates (citation tag {DEU} unresolved in sources table)
Frontier Model Forum Frontier Model Forum: Third-Party Assessments	“Safety case analysis as a form of holistic review (see C1 above).”	close confidence: high	Frontier Model Forum: Third-Party Assessments

Divergence

Where sources materially disagree

Added to the candidate list (ALIGNMENT-MATRIX.md §5) specifically because 'safety case' has an inconsistent operational status across sources: fully defined and instanced by Google DeepMind, used as an undefined trigger term by OpenAI and Meta (DU), not used as a label at all by Anthropic despite an equivalent claim-graph structure existing, and known only via discovery for UK AISI.