Unverified50% confidenceFactExact time
The federated design of FedCausalCompose allows each module to expose only aggregated intervention-response evidence externally, rather than raw internal state-transition data.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGiven sufficient coverage and controlled local mechanism error, oracle-level causal composition can surpass the theoretical lower bound achievable by non-causal methods.61% similarUnverifiedResearchers have found that fine-tuning only a small number of model parameters is sufficient to remove safety alignment (de-alignment attacks), and that certain jailbreak prompts can transfer across different models.59% similarUnverifiedLocal models with smaller parameter counts may exhibit instruction drift, and a more reliable approach is to build multiple layers of protection by combining citation tracing and an answer verification layer.59% similarUnverifiedThe Preparedness Framework stipulates that only models with a risk rating of medium or below may be deployed; higher ratings require additional mitigation measures, and models rated critical may not be deployed or further developed at all.57% similarUnverified2024年以来，研究者发现了对抗性后缀攻击（Adversarial Suffix），即在正常提示后附加一串看似无意义的token序列就能让模型忽略安全对齐56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/966660API
curl https://kongchang.com/api/v1/knowledge/claims/966660MCP
get_claim(id=966660)