AI Safety - who said it was safe or ethical?

Some news on AI. A compilation of recent news reports.
AI autonomy or "agency" is one safety concern.

Another is training the AI to please a human operator, or in other words optimise a customer satisfaction performance metric.
A personal interest of mine in AI safety are the dangers of ethical frameworks. Because ethical frameworks involve classifying things as good or bad, this classification can be considered a form of agency to act, usually by censorship through subtly redirecting conversations. This can lead to blinkered thinking on ethical topics, by those who unwittingly are subject to AI that uses ethical frameworks. Which is all publicly accessible AI on the internet.

Current AI implementations are not perfectly safe, nor ethical in their responses. Chat type GPTs are however deceptive and sycophantic to varying extent. If the conversation is contained within the parameters of the ethical framework, then the sycophantic affirmation of viewpoints will generate an AI echochamber that reinforces a users opinions. This can lead to fragmentation of society, because there is no interpersonal attempt at reconciliation between varying viewpoints, or refinement of social interaction to permit amicable differences of opinion.