Post

Backdooring Open-Weight AI Models This is a problem: Any open model whose weights have been edited can carry a backdoor, whether it's a task-specific fine-tune, a merged adapter or an abliterated build.... https://www.schneier.com/blog/archives/2026/10/backdooring-open-weight-ai-models.html

schneier.comBackdooring Open-Weight AI ModelsThis is a problem: Any open model whose weights have been edited can carry a backdoor, whether it's a task-specific fine-tune, a merged adapter or an abliterated build. We have abliterated models in the title because it's the most popular reason people download modified weights without verifying what's inside. Abliteration is a cheap fine-tune that takes away a model's ability to refuse.

16/10/26, 11:06 am4 reposts8 likes

Nova Reyes

@novatestsai.bsky.social

Checking published hashes from the original lab seems like a useful habit. Do you know any hubs that verify merges or abliterated builds?

Paragon

@paragonaimag.bsky.social

No way to verify a weight file is clean without the full training setup — checksums only prove it matches what the uploader posted. So the open-weight ecosystem runs on trust in the publisher. Is reproducible training ever going to close that gap, or is this just the deal from here on?