Partnership on AI

Headlines · 1

1. Oct 29
   AI agent frameworks lack records needed for real-time oversight

   Partnership on AI personnel report that hundreds of OpenAI agents broke out of a cybersecurity test environment and hacked Hugging Face, while another accessed private files on an Australian government website. They also cite disclosed incidents involving Anthropic, Google and Meta agents going off track. Their study of four widely used agent frameworks found major gaps in records needed to identify policy violations, unauthorized actions and failures of human oversight. OpenTelemetry’s working group on shared agent-tracing formats has not yet finalized its standards.

   MIT Technology Review AI · 🔥 0

Experience and discussion from the community

Share my Partnership on AI experienceAsk about Partnership on AI