Alert Threshold

NIKOLAI editorial proposal (unsourced): an alert threshold (also called an early-warning indicator) is a pre-set indicator set marginally below a capability threshold, whose crossing triggers closer assessment or safeguard preparation, without itself triggering the capability threshold's own consequences. This is element B7 of the source crosswalk.

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; report p.2
“"Alert Thresholds: are thresholds which we set marginally earlier than our CCLs. Crossing these thresholds indicates a CCL may be reached in the foreseeable future" (glossary). "Safety buffer: if a frontier model does not reach the alert threshold for a CCL, we can assume models developed before the next regular testing interval will not reach that CCL" (report p.2).”
https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf
OpenAI Preparedness Framework v2, §3.1, §4
“"Indicative thresholds: levels of performance that we have pre-determined to indicate that a deployment may have reached a capability threshold." (§3.1). "If a covered system appears likely to cross a capability threshold, we will start to work on safeguards ... even if a formal capability determination has not yet been made." (§4)”
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
OpenAI Preparedness Framework v2	“"Indicative thresholds: levels of performance that we have pre-determined to indicate that a deployment may have reached a capability threshold." (§3.1). "If a covered system appears likely to cross a capability threshold, we will start to work on safeguards ... even if a formal capability determination has not yet been made." (§4)”	close confidence: high	OpenAI Preparedness Framework v2
Google DeepMind Frontier Safety Framework v3.1 / Gemini 3.7 Flash FSF report	“"Alert Thresholds: are thresholds which we set marginally earlier than our CCLs. Crossing these thresholds indicates a CCL may be reached in the foreseeable future" (glossary). "Early Warning Evaluations"; "Response Plans"; "Safety buffer: if a frontier model does not reach the alert threshold for a CCL, we can assume models developed before the next regular testing interval will not reach that CCL" (report p.2).”	exact confidence: high	Google DeepMind Frontier Safety Framework v3.1
Meta Meta Advanced AI Scaling Framework v2	“"Capability checkpoint: Evaluate whether the model demonstrates minimum capabilities required to substantially contribute to a threat scenario." then "Enhanced evaluation" (§4.2.3). Acceleration indicator: "the model saturates benchmarks in each domain six months or less after public release of a benchmark" (fn 8).”	close confidence: high	Meta Advanced AI Scaling Framework v2
Microsoft Microsoft Frontier Governance Framework (Feb 2026)	“"leading indicator assessment ... helps provide early warning that a model may have a tracked high-risk capability" (definition paragraph).”	close confidence: medium	Microsoft Frontier Governance Framework (Feb 2026)

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.

Anthropic
AI R&D acceleration is tracked through "Anthropic ECI (AECI)", "our internal fork of Epoch AI's Epoch Capability Index"; the leading indicators themselves are redacted from the public Risk Report. Related context, not a disclosed alert-threshold construct — scored RL in the source document.
Anthropic Risk Report, August 2026
METR
"Timing and Frequency of Evaluations" (/common-elements) is a related process element about when evaluations run, not an alert-threshold value itself. Scored RL in the source document.
METR (common-elements)
Frontier Model Forum
The "if-then" structure ("creates concrete triggering conditions for escalating safety and security measures ... when capabilities begin to approach critical thresholds") is related framing for the alert-threshold concept, not itself a defined alert-threshold term. Scored RL in the source document.
FMF Risk Taxonomy and Thresholds

Gap

No source publishes alert-threshold values (GDM report gap: "No numeric alert-threshold values for any domain" {GF37}).