Verified85% confidenceFactExact time
Under the EU AI Act, "systemic risk" models trained with computing power exceeding 10^25 FLOPs are subject to additional safety assessments.
5
Sources
85%
Confidence
Long-term
Relevance
5/28/2026
First Seen
Sources
Related Entities
Related Claims
Unverified欧盟AI法案对系统性风险模型的阈值设定为10^25 FLOPs计算量，GPT-4级别模型适用82% similarUnverified《AI法案》对计算量超过10^25 FLOPs的超大规模GPAI模型要求进行对抗性测试并向主管机构报告严重事故74% similarUnverifiedThe WRI uses a multiplicative model: risk = exposure × (vulnerability + lack of coping capacity), with index values from 0 to 100, where higher values indicate greater overall risk.66% similarUnverifiedPopulation risk refers to the expected loss of a model over the entire data-generating distribution, while empirical risk is the average loss computed over a finite training sample.64% similarUnverified25%的风险降幅数字仍需在方法论严格的复制研究中验证，应被视为值得进一步验证的信号而非最终定论63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1965API
curl https://kongchang.com/api/v1/knowledge/claims/1965MCP
get_claim(id=1965)