Impact on AI Models

Anthropic-Led Study Finds 250 Poisoned Files Can Backdoor Large AI Models

Researchers say the fixed-count threat makes rigorous data provenance urgent.