You are viewing a single comment's thread from:RE: Qwen 3.8 27B Released - My observationsView the full contextView the direct parentthemarkymark (81)in #ai • 16 days ago I primarily only use local models and do not trust chinese hosted APIs.
Yes that's what I meant. Do local models have risk of that stuff being baked into its code? Or would have security researchers found it by now?
Is it possible? Yes.
Is it common? No.
As long as the weights are not malicious, local is completely safe. There is no phone home or anything like that.
It is possible to embed malicious intent into weights, you can even build weights into other weights that only activate under a trigger phrase.