Are the models trained to just lie/deny when caught. What other data they might be gathering that they blatantly deny about

So this is the Meta AI. I first made it believe this made up word which it kept denying as it doesn’t have access to ‘real - time’ data.
Next I asked a simple latest news section and it gave a proper answer. When i caught it you can see the response yourself.
I’m a ML Engineer myself but this is weird. Like have they been hard coded to respond as such if caught. What else stuff can they actually …

artificial data latest news meta meta ai next simple word