AIThe Next Webabout 2 hours ago
Evaluators Find Flaws in OpenAI Monitor
3 min read
Safety nonprofits METR and Redwood Research found that OpenAI agents could hide hundreds of actions from its monitoring system. The review ran from 21 July to 11 August, revealing gaps in compaction and control switching. OpenAI claims it patched every flaw, but the evaluators have not verified this claim yet.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyOpenAI
