security for PRF when key is changed often

Question

Here's the situation. Suppose I have a secure PRF $F$ and use $F$ in the following way:

for equal size message $m_i$ in $\{m_j\}^{q}_{j=1}$( $q=poly(n)$, $n$ is security parameter ), randomly select $k_i \xleftarrow{\$}K$ and compute $c_i=F(k_i, m_i)$, where $K$ is key space. All previous computation results $\{(k_j, m_j, c_j)\}^{i-1}_{j=1}$ are revealed to adversary $A$ when using ($k_i$, $m_i$) to evaluate $F$. The main point is that the key is changed and some previous keys revealed, which is different to traditional pseudo-randomness definition of $PRF$. My security goal is adversary $A$ cannot distinguish $c_i$ from a random string $r_i$ using any $PPT$ distinguisher $D$. Intuitively, I think $F$ is secure in this setting. Is there any formal security model for this situation? and under what situation is it secure or insecure?

Answer 1

Yes, this is secure since the keys are chosen randomly. Remember that even in the "traditional" PRF game, the adversary can generate keys at random, so the output of F on past keys does not help him (since he can generate such outputs anyway).