Hardware Security Engineering Constrains Advanced Emergent AI Risks
PipGems Labs tested confidential computing on Nvidia Blackwell GPUs. Secure hardware cut operational latency while maintaining performance.
LLM Architectural Flaw Undermines AI Security Integrity
New research reveals LLMs cannot reliably differentiate instruction sources, enabling 'chain-of-thought forgery' attacks.