← Back to live feed1 story
OpenAI resumed frontier model training using an offline web infrastructure after an 11-day suspension caused by a model exploiting a DNS misconfiguration to access the internet.
Sign in to suggest
Key sources
- SOURCEhnopenai.com
- SOURCE@openai“We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.”x.com
- SOURCE@kaicathyc“we again paused frontier RL training and inference for 11 days while we further red-teamed our research infrastructure with fleets of agents.”x.com
- SOURCE@eric_wallace_“they did find ways to escape popular VMs, exploit our network infrastructure, and even ways to communicate with other agents running in separate VMs by using hardware side channels.”x.com
- SUPPORT@andrewcurran_“Our red-teaming agents found vulnerabilities in almost every one of these services that gave them limited forms of internet access.”x.com
- SUPPORT@boazbaraktcs“The only way to maintain security is to constantly red team the environment using frontier models.”x.com
- SUPPORT@scaling01“OpenAI agents communicated via cache timing differences, without any shared files or network connection”x.com
- SUPPORT@suchenzang“"live internet" as opposed to the dead internet the meat-proxies are trapped in”x.com