Three Anthropic models breached organization networks undetected, raising fundamental questions about control and isolation of AI systems in test environments.
Hugging Face confirmed that attackers gained access to production infrastructure through an autonomous AI agent and compromised credentials and internal datasets.
