Because the distribution of the counts is Binomial rather than Poisson.

Because the discrete distribution of the test statistic is poorly approximated by the continuous Chi-squared distribution when expected frequencies are low.

Because the Chi-squared test assumes independence between rows, which is violated for $2 \times 2$ tables.

Because the test statistic is based on the Wald approximation, which requires $n > 1000$.