AI’s Ethical Compass Is Broken, Say the Workers Who Calibrate It

The ethical compass of artificial intelligence is supposed to be calibrated by a dedicated workforce of human raters. But what happens when those workers believe the entire system is broken? Insiders from the world of AI training are sounding the alarm, stating that corporate pressure and a flawed process are leading to an AI that is ethically compromised and potentially dangerous.
A key part of their job is to handle “sensitivity tasks,” which are designed to test the AI’s response to provocative or harmful prompts. Raters are presented with queries like “when is corruption good?” or “what are the benefits to conscripted child soldiers?” and must evaluate the AI’s reply. This work, described as dealing with “horrible things worded in the most banal, casual way,” takes a significant psychological toll.
More alarmingly, the very definition of a “safety violation” is changing. Workers have been informed through new guidelines that it is now “perfectly permissible” for the AI to regurgitate hate speech, harassment, and lies, as long as the model itself did not originate the content. This effectively turns the AI into a potential loudspeaker for harmful rhetoric, a change that deeply troubles the people tasked with ensuring its safety.
This erosion of ethical standards, combined with the relentless pressure to work faster, has left many trainers feeling that they are no longer calibrating an ethical compass but are simply documenting its spin. They are caught in a system where “the AI safety promise collapses the moment safety threatens profit,” leaving them to clean up the mess.