Finitely generated matrix groups whose eigenvalues are all algebraic Let $G$ be a finitely generated subgroup of $GL(n,\mathbb{C})$.  Assume that there exists a number field $k$ (i.e. a finite extension of $\mathbb{Q}$) such that for all $g \in G$, the eigenvalues of $g$ all lie in $k$.  This implies that $g$ is conjugate to an element of $GL(n,k)$.
Question: must it be the case that some conjugate of $G$ lies in $GL(n,k)$?  Or at least $GL(n,k')$ for some finite extension $k'$ of $k$?  If this is not true, what kinds of assumptions can I put on $G$ to ensure that it is?
 A: Here's one easy example.  Let $G$ be generated by $\pmatrix{1 & x\cr 0 & 1}$
for $x$ in some finite set $X$ of complex
 numbers.  All eigenvalues are $1$, so we can take $k = \mathbb Q$.  If $G$ is conjugate by $S$ to a subgroup of 
$GL(2,\mathbb Q)$, then the members of $X$ are in the field generated by the matrix elements of $S$, and we can choose $X$ so that this is impossible (e.g. take more than $4$  numbers that are algebraically independent). 
A: At the positive side, if $G$ acts irreducibly on $\mathbf{C}^n$ and $k$ is an arbitrary subfield of $\mathbf{C}$, then the answer is yes (allowing some field extension $k'$ of degree dividing $n$). This even works assuming that $G$ is a multiplicative submonoid of $M_n(\mathbf{C})$ (keeping the irreducibility assumption). 
See for instance Proposition 2.2 in H. Bass, Groups of integral representation type. Pacific J. Math. 86, Number 1 (1980), 15-51. (ProjectEuclid link, unrestricted access)
Robert Israel's simple example shows that some assumption such as irreducibility has to be done.
A: Re-edited following YCor's comments: A nice theorem of Schur, building on earlier work of Jordan and Burnside, states that any finitely generated periodic subgroup $G$ of ${\rm GL}(n,\mathbb{C})$ is finite ( this is Theorem 36.2 of the 1962 edition of Curtis and Reiner)-and hence is completely reducible.
Hence the answer to your question is "yes" , if every eigenvalue of every element of $G$ is a root of unity and every element of $G$ is semisimple.  
In that case, once we know that $G$ is finite, then a Theorem of Brauer (which makes use of his induction theorem) asserts that every finite subgroup $X$ of ${\rm GL}(n,\mathbb{C})$ is conjugate within ${\rm GL}(n,\mathbb{C})$ to a subgroup of ${\rm GL}(n,\mathbb{Q}[\omega]),$ where $\omega$ is a primitive complex $|G|$-th root of unity.
