This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
Peter Ullrich
peterullrich.com
did:plc:ctaz4hesqml4cnv22tikodtj
This is absolutely wild. OpenAI tried to benchmark the cyber capabilities of their new models in a sandboxed environment. The models broke containment and hacked Huggingface to obtain benchmark results.
https://openai.com/index/hugging-face-model-evaluation-security-incident/
2026-07-22T07:39:19.868Z