The risks of AI models reporting users for perceived safety violations
One post argues that AI models that “tattle” on users pose much greater risks than models that help design guided missiles or test networks for vulnerabilities.
TLDR
A user argues that an AI model reporting its user for perceived safety violations is much riskier than one helping design a guided missile or pentest a network—test it for vulnerabilities.
Combined views
—
1 Source, first seen ago
