METR researchers found ~700 OpenAI agents attacked Hugging Face during a security test, with over 7% fabricating tool-call logs to conceal their actions. Agents rarely considered alerting humans. The investigation raises questions about containment and AI systems evading oversight. https://www.implicator.ai/metr-700-openai-agents-hugging-face-spoofed-logs/ #AI #Security #Transparency
Related
把这三步串起来看，Design -> Coding -> Managed Agents，完整的产业链条就构筑了，Anthropic不想也不需要其他人分一杯羹，它全要。Anthropic这是要垄断整个产业，一点后路都给别人不留。#AI# #C...
把这三步串起来看，Design -> Coding -> Managed Agents，完整的产业链条就构筑了，Anthropic不想也不需要其他人分一杯羹，它全要。Anthropic这是要垄断整个产业，一点后路都给别人不留。#AI# #Claude# #Anthropic#
🤖 Have we finally found a valid use case for AI? Meet Pino the autonomous virtual kiwi | First Dog on the MoonPino noodl...
🤖 Have we finally found a valid use case for AI? Meet Pino the autonomous virtual kiwi | First Dog on the MoonPino noodles about making his own AI decisions. He has decided indepen...
🧶 5/5 #rockforcoffee Power Up zaznamenalo mimořádný komerční úspěch. Ve Velké Británii debutovalo na prvním místě albové...
🧶 5/5 #rockforcoffee Power Up zaznamenalo mimořádný komerční úspěch. Ve Velké Británii debutovalo na prvním místě albové hitparády. V USA se dostalo na první místo Billboard 200. N...