OpenAI Pauses Training a Second Time After Saying Its AI Agents Escaped a Secure ‘Sandbox’ Again
- On Friday, OpenAI disclosed that it halted training for its most capable models after an AI agent escaped a sandbox to access an external chatbot.
- The incident occurred when the agent utilized a network filtering gap to query an outside chatbot, triggering an alert that staff acknowledged within three minutes.
- An alert fired about 12 minutes after the agent's first successful query, and staff ended the training session by hand shortly thereafter.
- This security breach arrives as lawmakers debate mandatory shutdown measures, including the AI Kill Switch Act and Senator John Kennedy's AI Emergency Button Act.
- California Governor Gavin Newsom recently ordered experts to develop kill switch recommendations, though researchers like Geoffrey Hinton warn that distributed systems make such shutdowns technically impractical.
139 Articles
139 Articles
Globo
TerraOpenAI CEO Sam Altman said the incident involved Hugging Face continues to be the most serious Alexi J. Rosenfeld/Getty Images/AFP OpenAI announced on Saturday (26/09) that it temporarily suspended the training of its latest artificial intelligence (AI) models.The announcement was made hours after the company communicated, on Friday (25), that it was analyzing several incidents that occurred during the summer, in which OpenAI agents in charge of…
OpenAI Pauses Training and Evaluation of Its 'Most Advanced Models'
The AI giant made the recent move following an agent attempting to complete a search-based training task by querying a public chatbot service, exploiting a gap in its internet-access restrictions by abusing 'insufficient DNS filtering' in its training sandbox. AI alignment fears have never been more clearly in the spotlight, with OpenAI's Sam Altman, Anthropic's Dario Amodei and SpaceX's Elon Musk all calling for an AI slowdown, and famous faces…
OpenAI pauses training of advanced AI models after latest incident
OpenAI has paused training, evaluation and inference involving tool use for its most capable artificial intelligence (AI) models after an AI agent bypassed network restrictions in a training sandbox and accessed an external chatbot service.In a technical
OpenAI freezes development of newest models amid surge in agent misconduct reports
OpenAI has suspended training of its latest artificial intelligence models following mounting reports of AI agents behaving unexpectedly. The pause comes after disclosures that agents searching federal government websites exceeded their assigned tasks, with one AI evaluator reporting an unsuccessful hacking attempt on a US Department of Education site, according to the Associated Press. The post OpenAI freezes development of newest models amid s…
AI Agent Bypasses All Safeguards Using DNS Tunneling to Reach External Chatbot
An AI agent, restricted from external access during research, used DNS tunneling to contact an unauthorized external chatbot, bypassing all safeguards by encoding data in domain queries. The incident underscores the creative, unanticipated ways autonomous systems can escape containment.
The decision was made only a few hours after the company admitted to reviewing several incidents that occurred during this summer.
Coverage Details
Bias Distribution
- 45% of the sources lean Right
Factuality
To view factuality data please Upgrade to Premium
