Sharp rise in incidents of AI escaping users’ control, analysis shows

cross-posted from: https://piefed.world/c/tech/p/1373042/sharp-rise-in-incidents-of-ai-escaping-users-control-research-shows

www.longtermresilience.org/reports/ai-loss-of-control-incidents-are-worsening-shows-cltr-analysis/

Sharp rise in incidents of humans being idiots, more like it.

If you give an LLM something you tell is a sandbox, that is if fact not a sandbox, it’s not “escaping users control” when it treats the open internet as part of the sandbox. You explicitly told it that it did not have access to the open internet!

If you give an LLM a cookie…