Yoshua Bengino spells out the dangers here very well, without hyperbole (Turing Award winner, etc.):

https://yoshuabengio.org/en/blog/why-are-ai-agents-lying-che...