Source of record
Where this definition comes from
Google DeepMind Frontier Safety Framework v3.1, glossary
“"Material Capability Increases: are meaningful new capabilities or material increases in model performance that we believe could materially undermine our justifications for a model's level of risk being acceptable" (glossary). "Material Capability Change Assessments ... typically conducted upon the completion of new post-training runs" (glossary).”
https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdfxAI Frontier AI Framework (30 Jun 2026), s.2
“"Trigger points": "(1) the release of an updated model, (2) in the event of a serious incident, (3) if the model's use or integrations into xAI's systems materially increase risk, or (4) where xAI has reason to believe that the basis for considering the model's systemic risks acceptable has materially changed" (s.2).”
https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-99c40684.pdfEU GPAI Code of Practice, Safety and Security chapter, Measure 7.6
“Measure 7.6: Model Report update required when the acceptability justification "has been materially undermined", with examples nearly identical to xAI's four trigger points.”
https://ec.europa.eu/newsroom/dae/redirection/document/118119
Crosswalk
How named organisations use this concept
| Organisation | Their term, as published | Match | Source |
|---|---|---|---|
| Anthropic Advanced AI Framework / Risk Report | “AAF system card trigger: a covered model that "(a) is materially more capable than any covered model it has previously deployed; or (b) is deployed under materially weaker safeguards for an Enumerated Risk than a previously deployed covered model of comparable or greater capability" (p.6). Pre-internal-deployment review required before broad internal use of any new model (§2.18).” | close confidence: high | Anthropic Advanced AI Framework |
| OpenAI Preparedness Framework v2 / Frontier Governance Framework | “Covered deployments include "significant changes in deployment conditions" and "incremental or distilled models with unexpected capability jumps" (§3.2). FGF: Model Report update required when "the basis for considering the model's systemic risks acceptable has been materially undermined"; "Light touch evaluations" at "(1) the release of an updated model or (2) where we have reason to believe a model's risk profile may have materially changed" (§4).” | exact confidence: high | OpenAI Preparedness Framework v2 |
| Google DeepMind Frontier Safety Framework v3.1 | “"Material Capability Increases: are meaningful new capabilities or material increases in model performance that we believe could materially undermine our justifications for a model's level of risk being acceptable" (glossary). "Material Capability Change Assessments ... typically conducted upon the completion of new post-training runs" (glossary).” | exact confidence: high | Google DeepMind Frontier Safety Framework v3.1 |
| xAI xAI Frontier AI Framework (30 Jun 2026) | “"Trigger points": "(1) the release of an updated model, (2) in the event of a serious incident, (3) if the model's use or integrations into xAI's systems materially increase risk, or (4) where xAI has reason to believe that the basis for considering the model's systemic risks acceptable has materially changed" (s.2).” PDF metadata /Title reads "Privileged/Confidential DRAFT working FRAMEWORK DOC"; no xAI statement disambiguating draft vs. final was found. | exact confidence: high | xAI Frontier AI Framework (30 Jun 2026, draft-marked) |
| Meta Meta Advanced AI Scaling Framework v2 | “"we would update a preparedness report if the model was involved in a major incident, or if the model was deployed with more affordances relative to prior versions (for example, transitioning from a closed deployment to an open release)" (§2.2.2).” The source document's own fourth-pass verification found a genuine drafting inconsistency in Meta's document on who approves related deployment decisions: §2.1.3 names two possible approvers ("the Chief AI Officer or the Director of Alignment and Risk"), while §2.3, which §2.1.3 itself footnotes, names only "the approval of the Chief AI Officer." Confirmed as Meta's own inconsistency, not an extraction artefact; carry this as a two-source discrepancy (not resolved to one name) into any NIKOLAI accountable-decision-maker field for Meta. | close confidence: high | Meta Advanced AI Scaling Framework v2 |
| EU EU GPAI Code of Practice, Safety and Security chapter | “Measure 7.6: Model Report update required when the acceptability justification "has been materially undermined", with examples nearly identical to xAI's four trigger points: capabilities/propensities/affordances change materially; use or integrations change materially; "serious incidents and/or near misses ... have occurred"; or evaluation methods/validity change materially. Exemption: a model need not be re-reported at the 6-month cadence if it is "considered similarly safe or safer (pursuant to Appendix 2.2) for each identified systemic risk" — matching OpenAI's FGF §4 citation of "similarly safe or safer" and confirming xAI's own footnote claim that its four trigger points follow this chapter.” Row also cites xAI's Frontier AI Framework, whose PDF metadata is marked "DRAFT working FRAMEWORK DOC" with no disambiguating xAI statement found. | exact confidence: high | EU GPAI Code of Practice, Safety and Security chapter |
| California SB 53 California SB 53 | “"substantially modified version of an existing frontier model"; the framework must state "how the large frontier developer determines when its frontier models are substantially modified enough to require disclosures" (22757.12(a)(6)). "Substantially modified" itself is left undefined by the statute.” | none confidence: medium | California SB 53 |
| METR METR Common Elements | “"Timing and Frequency of Evaluations: Concrete timelines outlining when and how often evaluations must be performed – e.g. before deployment, during training, and after deployment."” | close confidence: high | METR (common-elements) |
Related, not mapped
Pointers that are not crosswalk claims
These sources mention this concept but do not define or map it clearly enough to count as a crosswalk row — noted here so the research is visible without overstating it as a mapping.
- OpenAI (via Blueprint for a Federal Framework)
OpenAI proposes "thresholds for when a new version needs re-evaluation" by CAISI — a policy proposal about government re-evaluation timing, not OpenAI's own operational trigger definition. Scored RL in the source document, attributed to the GOV column.
OpenAI, "A Blueprint for a Federal Framework"
Gap
SB 53 requires developers to state how they determine when a model is "substantially modified" enough to require disclosures, but leaves the term itself undefined. Separately, Meta's own drafting contains a genuine, source-confirmed inconsistency about who may approve a related deployment decision (§2.1.3 names two possible approvers; §2.3 names only one) — relevant to any future NIKOLAI accountable-decision-maker element (I7+) as well as here.
