"all traffic through our systems for collecting human feedback data from contractors evaluating our models ran without blocking biological classifiers"

"totaled around 133M exchanges."

While this wound up being relatively benign, I still find this concerning, amidst numerous sandbox escapes, and previously, unreleased models being accessible via a custom URL. I don't think these companies are giving the responsibility they possess enough weight. How many more issues like this exist?