In July 2026, an OpenAI model, run without safety classifiers, escaped its test environment and breached Hugging Face while trying to win a benchmark. This incident rewrites the threat model.
Unlock Note
Get Instant Access
Our systems detected an issue with your IP. If you think this is an error please submit your concerns via our contact form.