You're taking cohomology of classifying spaces, aren't you? I thought first that the answer was positive, but there was a mistake in my argument. I can recycle the argument to show that you obtain a counterexample whenever the following two conditions hold: $$\hom(\pi_1G,A)^{\pi_0G}\neq 0,\qquad H^2(BG,A)=0.$$ I don't have any example satisfying these conditions off the top of my head, but I share this in case someone comes up with one (maybe $O(2)$ with some local coefficients $A$?) or shows it's impossible. The short exact sequence $$G_c\hookrightarrow G\twoheadrightarrow G/G_c=\pi_0G$$ induces a fibration after taking classifying spaces $$BG_c\hookrightarrow BG\twoheadrightarrow B\pi_0G.$$ This, in turn, gives rise to a Serre spectral sequence $$E_2^{p,q}=H^p(B\pi_0G,H^q(BG_c,A))\Longrightarrow H^{p+q}(BG,A).$$ The spectral sequence differential $d_r$ has bidegree $(r,1-r)$ and the cohomology groups $H^q(BG_c,A)$ are regarded here as $\pi_0G$-modules. The morphism whose infectivity you want to prove decomposes as $$E_2^{3,0}=H^3(B\pi_0G,A)\twoheadrightarrow E_\infty^{3,0}\hookrightarrow H^3(BG,A).$$ We have three exact sequences $$E_2^{1,1}\stackrel{d_2}{\longrightarrow}E_2^{3,0}\twoheadrightarrow E_3^{3,0}$$ $$E_3^{0,2}\hookrightarrow E_2^{0,2}\stackrel{d_2}{\longrightarrow}E_2^{2,1}$$ $$E_\infty^{0,2}=E_4^{0,2}\hookrightarrow E_3^{0,2}\stackrel{d_3}{\longrightarrow}E_3^{3,0}\twoheadrightarrow E_4^{3,0}=E_\infty^{3,0}$$ The group $$E_2^{n,1}=H^n(B\pi_0G,H^1(BG_c,A))$$ vanishes because $G_c$ is connected hence $BG_c$ is simply connected so $H^1(BG_c,A)=0$. Therefore $$E_3^{3,0}=E_2^{3,0},$$ $$E_3^{0,2}=E_2^{0,2}=H^2(BG_c,A)^{\pi_0G}=\hom(\pi_1G,A)^{\pi_0G}.$$ If $H^2(BG,A)=0$ then $E_\infty^{0,2}=0$ and $d_3\colon E_3^{0,2}\rightarrow E_3^{3,0}$ must be injective. This spectral sequence differential is precisely the inclusion of the kernel of the map whose infectivity you wanted.