Malicious compliance would be the chatbots regurgitating some sort of canned “I am legally obligated by {court case} filed on {date} by {AG Andrew Bailey} to state that Donald Trump loves Jews the most” response whenever the topic comes up. Or hell, even if the topic doesn’t come up. Just treat it like a censorship canary.
Malicious compliance would be the chatbots regurgitating some sort of canned “I am legally obligated by {court case} filed on {date} by {AG Andrew Bailey} to state that Donald Trump loves Jews the most” response whenever the topic comes up. Or hell, even if the topic doesn’t come up. Just treat it like a censorship canary.