There is no such thing as "the" probability without more input. You can calculate a number with Bayesian methods by making the following additional assumptions:
- First assuming a prior probability distribution over the possible values of $p$. The simplest option is the uniform distribution on $[0, 1]$. If $n$ is large enough it doesn't matter too much which prior you pick.
- Then assuming that each of your experiments is independent. This is a strong assumption and in my opinion it's quite debatable whether it holds in this case, at least without putting in a certain amount of work to make sure that your experiments are "sufficiently different" from each other.
With those additional assumptions we can use Bayes' theorem to update on the observation that $n$ independent experiments don't produce the bug. The posterior probability distribution over the possible values of $p$ is then proportional to the likelihood $(1 - p)^n$, which gives
$$\mathbb{P}(p \le x) = \frac{\int_0^x (1 - t)^n \, dt}{\int_0^1 (1 - t)^n \, dt} = \boxed{ 1 - (1 - x)^{n+1} }.$$
The median of this distribution can be computed by setting this probability to $\frac{1}{2}$, which gives $(1 - x_{\text{median}})^{n+1} = \frac{1}{2}$ or
$$x_{\text{median}} = 1 - \sqrt[n+1]{\frac{1}{2}} = 1 - \exp \left( - \frac{\log 2}{n+1} \right) \approx \frac{\log 2}{n+1}.$$
So this distribution is concentrated around $O \left( \frac{1}{n} \right)$ which should be reasonably intuitive. You can also compute the posterior mean using the Beta function integral, which gives
$$\mathbb{E}(p) = \frac{\int_0^1 t(1 - t)^n \, dt}{\int_0^1 (1 - t)^n \, dt} = \frac{1}{n+2}.$$
This is a special case of Laplace's rule of succession.
