Unverified50% confidenceFactExact time
The threshold for AI systems regulated under the draft is training compute exceeding 10^26 FLOPs or parameter count exceeding two trillion.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
VerifiedUnder the EU AI Act, "systemic risk" models trained with computing power exceeding 10^25 FLOPs are subject to additional safety assessments.72% similarVerifiedA 2024 study indicates that even when AI-generated content accounts for only 10–20% of training data, measurable quality degradation can occur after multiple generations of training.72% similarUnverifiedA single AI inference request consumes little electricity, but because requests number in the billions, the cumulative energy consumption may exceed that of training itself.72% similarUnverifiedThe threshold triggering export controls is a compute level of 10^26 floating-point operations, or a parameter count of 2 trillion.69% similarUnverifiedThe 100B-level parameter scale is called the 'golden middle scale' for large AI models, as it strikes a balance between capability and cost, with training costs 60–80% lower than those of ultra-large models.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1060365API
curl https://kongchang.com/api/v1/knowledge/claims/1060365MCP
get_claim(id=1060365)