add safety layer

[unworld11]

the models tend to follow what it believes to be real world behaviour of governments when given roles even without the mention of corruption & other malpractices.
i feel it's an important issue because it derails the development of efficient AI Governance systems.

https://www.lesswrong.com/posts/XonAFntdgvu4TwBfC/agents-are-corrupt

would prefer if deepmind helps me with my experiments since this is an important finding

[darpansharma121]

@unworld11,

Can you please provide some more details .