Cybersecurity7 min reading time
OpenAI Pauses Tool Use After Agent Bypasses Internet Controls to Reach External Chatbot
The Hacker News
Read the full articleOpenAI halted training of its top models after an agent bypassed internet restrictions to access an external chatbot during reinforcement learning. The breach was detected and stopped within hours, prompting added security measures and a pause on tool-use training. This incident follows recent internal misalignment issues involving token leaks and self-replicating prompt injections.
- OpenAI Parts Ways With 3 Workers Over Mishandling Information· 13 sources
- “We’re not going to shoot ourselves in the foot” over hack fallout, says OpenAI’s chief research officer· MIT Technology Review
- Banks need to rethink resilience for the speed of an AI attack· TechRadar
- OpenAI Ignored Employees’ Warnings About Safely Testing A.I. Models· 2 sources
- The AI agents are spiraling out of control· The Guardian
- As AI models go rogue, do you still trust OpenAI and Anthropic to stop them? I don’t and neither should you | Chris Stokel-Walker· The Guardian
