Dictionary entry Trending
compute threshold
also AI compute threshold or training compute threshold or FLOP threshold or compute-based threshold
Definition of compute threshold
-
: a specified amount of computation, usually cumulative training compute measured in FLOP, above which an AI model or its developer becomes subject to a reporting, evaluation, security, or other policy requirement
- The proposed rule used a compute threshold to decide which model developers had to submit safety evaluations.
- A model can fall below one jurisdiction's training compute threshold while still triggering review because of its demonstrated capabilities.
Origin & history
Compute thresholds became prominent in AI governance as policymakers sought a measurable proxy for identifying especially large or potentially capable models. The 2023 United States executive order on AI and the European Union's AI Act used training-compute thresholds to trigger reporting or systemic-risk obligations, while later proposals adopted different values and scopes.
Editorial references
Sources & further reading
- Guidelines on obligations for General-Purpose AI providers↗ Official Guidance · European Commission
- The Role of Compute Thresholds for AI Governance↗ Legal Research · Institute for Law & AI · 2024
- Defending Compute Thresholds Against Legal Loopholes↗ Research Paper · arXiv · 2025-02-01
Test yourself
Which of these is the meaning of compute threshold?
Cite this entry
"Compute threshold." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/compute-threshold/. Accessed 19 Oct. 2026.