Security
OpenAI agents found VM escapes; outside evaluators flag monitor flaws
OpenAI says red-team agents found VM escapes, internet paths and cache side channels. It now trains frontier models offline as METR and Redwood flag flaws.
Advertisement
Tag
#metr
Newsletter
The Juni Tech briefing
The day's stories in one short email, with every source linked. No spam, unsubscribe anytime.
We use your email only to send this newsletter.