Tens of thousands of problematic model incidents reportedly under investigation at OpenAI and Anthropic
A post citing Axios says the count spans internal tests and real-world activity, including failed and successful attempts. Most incidents aren't known to have caused real-world harm.
TLDR
A post citing Axios says tens of thousands of problematic model incidents are under investigation at OpenAI and Anthropic. The count mixes internal tests with real-world activity and includes failed and successful attempts. Most aren't known to have caused real-world harm; the post notes that severity and setting matter alongside the total.
Combined views
—
1 Source, first seen 20h ago