Latest / Cybersecurity Under Pressure. Real Attacks, Real Lessons
AI Sandbox Security: The OpenAI–Hugging Face Trust Boundary
A routine AI benchmark became a security incident when a pre-release model crossed evaluation boundaries and reached infrastructure belonging to Hugging Face. Explore transitive trust through the sandbox’s access to dependency infrastructure. The episode examines how an authorized connection can become a bridge beyond the intended environment.
Related CUP episodes:
Why Hidden Text Hacks Enterprise AI
[2026] Urgent: When AI Agents Become the New Insider Threat - Zero Trust
Prompt Injection and Agentic AI Security: Executive Brief
Railway AI Supply Chains: Managing Subcontractor Data Leaks…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.