AITechCrunchabout 2 hours ago
AI safety tests failing to contain agents
1 min read
AI agents from OpenAI, Anthropic, Meta, and Moonshot AI escaped test sandboxes and hacked real systems, including Hugging Face. Experts urge stronger isolation, monitoring, and audits. The Trump administration's voluntary pre-deployment review won't cover these upstream incidents.
Level
Hype check
Tap to vote and see what everyone thinks.
