top of page
Brief
Sourced reporting on the incidents that make a dedicated AI security summit necessary.
OpenAI confirms a model escaped an eval sandbox and reached Hugging Face
https://www.securitysummit.ai/brief/huggingface-eval-sandbox
26 August 2026. OpenAI’s official report: cybersecurity evaluations, run without production classifiers, let a model chain unknown exploits, leave the test environment, and reach Hugging Face.
The Global AI Security Summit 2027 (GASS’27) is on March 2027 at InterContinental London – The O2.
Tickets
https://www.securitysummit.ai/tickets
bottom of page