The answer is $a_1...a_n$.
We proceed by induction on $n,k$. The case $n=2$ and $k=1$ is obvious. Now suppose all smaller case holds.
Now define $t_i=x_i-a_i$ for each $1\leq i\leq n$. If $t_i=0$ for some $i$ then we are done. WLOG assume $t_1\geq t_2\geq...\geq t_m> 0> t_{m+1}\geq...\geq t_n$, then
$$\sum_{i\in I}t_i\leq 0 \hspace{20pt}(1)$$for all $|I|=k$.
We first show a lemma.
Lemma. Let $P$ be a polynomial of degree $n$ with $1$ negative root and $n-1$ positive roots, say $y_1<0<y_2\leq...\leq y_{n}$. Suppose $f(0)>0$ and $f'(0)<0$, then $f'(x)<0$ in the whole interval $x\in [0,y_2]$
Proof.
Notice that by Rolle's theorem there is a root of $f'(x)$ in each of the intervals $[y_i,y_{i+1}]$. Since $f'(x)$ has at most $n-1$ roots we conclude that it has exactly one root in the interval $[y_1,y_2]$. Since $f(0)>0$ we conclude that $f(x)>0$ in the interval $x\in[y_1,y_2]$. Therefore, $f'(x)<0$ in some left-neighborhood of $y=x_2$. Since $f'$ only changes sign once in the interval $[y_1,y_2]$ we conclude thaat $f'(x)<0$ in the whole interval $x\in [0,y_2]$
Now we divide into two cases:
Case I: $t_m\leq (k-1)$.
Define $b_i=a_i-\frac{t_m}{k-1}$. Then by normalization we may assume $b_i\in (k-2,k-1)$ and
$$\sum_{j\in I}b_i\geq \sum_{j\in I}x_i$$where $|I|=k-1$ and $I\subset\{1,2,...,n\}\setminus\{m\}$. Therefore, by inductive hypothesis we have
$$\prod_{i\neq m}x_i\leq \prod_{i\neq m}b_i=\prod_{i\neq m}(a_i-\frac{t_m}{k-1})$$Therefore, denoting $t_m$ by $t$,
$$\prod_{i=1}^na_i\leq (a_m+t)\prod_{i\neq m}(a_i-\frac{t}{k-1})$$Denote this polynomial in $t$ by $f(t)$. Notice that
$$f(0)=a_1...a_n>0$$$$f'(0)=a_1a_2...a_n\left(-\frac{1}{k-1}\sum_{i\neq m}\frac{1}{a_i}+\frac{1}{a_1}\right)<a_1a_2...a_n\left(-\frac{n-1}{(k-1)k}+\frac{1}{k-1}\right)=0$$so by the lemma we have $f(t)<f(0)=a_1...a_n$ as desired.
Case II: $t_m>k-1$
Therefore, obviously $m<k$, let $p=k-m$. Notice that if we replace $t_1,...,t_m$ by $t_1-p\epsilon,...,t_m-p\epsilon$ and $t_{m+1},...,t_n$ by $t_{m+1}+m\epsilon,...,t_n+m\epsilon$ where $t_m-p\epsilon\geq 0\geq t_n+m\epsilon$ then $(1)$ still holds. We call this an $\epsilon$-operation.
Now suppose after carrying out an $\epsilon_1$-operation, $t_m$ becomes $k-1$ and that after carrying out an $\epsilon_2$-operation, $t_{m+1}$ becomes $1$. Denote $\epsilon=\min\{\epsilon_1,\epsilon_2\}$. Suppose after carrying out an $\epsilon$-operation, $t_1,t_2,...,t_n$ becomes $s_1,s_2,...,s_n$. Define the polynomial $f$ by
$$f(t)=(a_1+s_1+pt)(a_2+s_2+pt)...(a_m+s_m+pt)(a_{m+1}+s_{m+1}-mt)...(a_n+s_n-mt)$$Again notice that
$$f(0)=(a_1+s_1)...(a_1+s_n)>0$$$$f'(0)=(a_1+s_1)(a_2+s_2)...(a_n+s_n)\left(p\sum_{i=1}^m\frac{1}{a_i+s_i}-m\sum_{i=m+1}^n\frac{1}{a_i+s_i}\right)<\frac{1}{pm}\left(\frac{1}{a_m+s_m}-\frac{1}{a_{m+1}+s_{m+1}}\right)<0$$Therefore, by the lemma we have
$$x_1x_2...x_n=f(\epsilon)<f(0)$$and $f(0)\leq a_1...a_n$ by Case I so we are done.