AI R&D Acceleration Criterion: Definition

AI R&D Acceleration Criterion

Definition

An AI R&D acceleration criterion is a measurable test for when AI progress has sped up enough, plausibly because AI is automating AI research, to count as crossing a capability threshold. Examples include progress at double the historical rate, or new benchmarks saturating within six months of release.

Also called: Dramatic acceleration, Double the rate of progress, ML R&D acceleration

NIKOLAI's proposed operational definition

NIKOLAI proposal: an AI R&D acceleration criterion is the measurable test a framework uses to decide that AI progress has accelerated beyond its historical rate, plausibly because AI systems are contributing to AI research. Record the measured quantity (aggregate capability, benchmark saturation speed), the baseline and window, the multiple or speed that counts as acceleration, and the attribution condition.

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

If baseline progress involved a 3× scaleup in compute and a 3× improvement in algorithmic efficiency (a 9× effective scaleup), double the rate of progress would mean something like an 81× effective scaleup (Anthropic RSP v3.4, section 1, fn. 8)
Meta's test requires benchmark saturation within six months of a benchmark's release, observed for at least two of the latest benchmarks in a domain within 18 months (Meta Advanced AI Scaling Framework v2, fn. 8)
Google DeepMind recommends Security level 3 for models at its ML R&D acceleration level 1 (Frontier Safety Framework v3.1, Table 3.2.2.a)

Counter-examples

Not to be confused with

Researcher productivity gains are not an acceleration criterion: Anthropic's policy notes that doubling inputs does not necessarily double the rate of progress.

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.

Anthropic ECI (AECI): Anthropic ECI (AECI) is Anthropic's internal fork of Epoch AI's Epoch Capability Index, used in its Risk Report to track whether AI progress is accelerating. It uses internal benchmark results, so it is not comparable with Epoch's public leaderboard.; Source: Anthropic Risk Report, August 2026

Source of record

Where this definition comes from

Anthropic Responsible Scaling Policy v3.4, Section 1 table, p.9, fn. 8-9
“"We would consider scenario (2) to have occurred where (a) we observe or expect double the rate of progress ... in AI aggregate capabilities compared to both the rate we'd expect and the fastest rate of extended progress ... we've observed in the absence of significant AI contributions to AI R&D"; fn. 8: "“Double the rate of progress” means “as much progress in one year as one would see in two years at baseline.”"; fn. 9: "Over at least three model generations."”
https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf
Meta Advanced AI Scaling Framework v2, fn. 8, p.35
“"(a) the model saturates benchmarks in each domain six months or less after public release of a benchmark; and (b) this acceleration pattern is observed consistently for at least two of the latest benchmarks released in that domain within an 18-month period."”
https://ai.meta.com/static-resource/Meta_Advanced-AI-Scaling-Framework-v2

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.

Organisation	Their term, as published	Match & verification	Source
AnthropicShadow mapping Responsible Scaling Policy v3.4	“"We would consider scenario (2) to have occurred where (a) we observe or expect double the rate of progress ... in AI aggregate capabilities compared to both the rate we'd expect and the fastest rate of extended progress ... we've observed in the absence of significant AI contributions to AI R&D" (p.9). fn. 8: "“Double the rate of progress” means “as much progress in one year as one would see in two years at baseline.”" fn. 9: "Over at least three model generations."”	exactEQ confidence: high	Anthropic Responsible Scaling Policy v3.4
MetaShadow mapping Meta Advanced AI Scaling Framework v2	“AI progress substantially accelerating from historical rates, operationalised as: "across multiple key capabilities domains (e.g., mathematical reasoning, scientific knowledge, multimodal reasoning, agentic task completion): (a) the model saturates benchmarks in each domain six months or less after public release of a benchmark; and (b) this acceleration pattern is observed consistently for at least two of the latest benchmarks released in that domain within an 18-month period." (fn. 8, p.35)”	exactEQ confidence: high	Meta Advanced AI Scaling Framework v2
Google DeepMindShadow mapping Frontier Safety Framework v3.1	“"ML R&D acceleration level 1: Has been used to accelerate AI development, resulting in AI progress substantially accelerating from historical rates." (Table 3.2.2.a)” Names the acceleration condition as a Critical Capability Level but publishes no measure or multiple.	closeCL confidence: high	Google DeepMind Frontier Safety Framework v3.1

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

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
The August 2026 Risk Report tracks acceleration with the Anthropic ECI (AECI), "our internal fork of Epoch AI's Epoch Capability Index"; the leading indicators themselves are redacted from the public version.
Anthropic Risk Report, August 2026
OpenAI
OpenAI's High threshold for AI Self-improvement is framed as researcher productivity (equivalent to giving every researcher a mid-career research engineer assistant), not as a rate-of-progress test (Preparedness Framework v2, Table 1).
OpenAI Preparedness Framework v2

Divergence

Where sources materially disagree

Anthropic measures the rate of aggregate capability progress against a historical baseline and requires attribution to AI R&D automation; Meta measures how quickly new benchmarks saturate; Google DeepMind states the condition without a measure; OpenAI frames the nearby threshold as researcher productivity. Anthropic's policy explicitly distinguishes the two framings: "This is not the same idea as “doubling researchers' productivity,” since doubling inputs does not necessarily double the rate of progress."

Gap

No source publishes the baseline values its criterion is measured against; Anthropic's AECI indicators are redacted from the public Risk Report, and Meta does not name the benchmarks it would count.