If the Vcc at the right side of R2 is in the allowed ADC input voltage range, there's no damage expected. The 2,5kOhm limit is for accuracy. the ADC takes a little input current and that can cause errorneous ADC result when compared to unloaded voltage. You said "it works" and I believe it - obviously you can take into the account somehow the possible conversion error or the exact value isn't interesting, you measure only if the voltage is over or under a certain treshold.
ADD due the comments:
It's difficult to decide how much more ADC input catches random noise due the bigger source impedance. Only careful tests can reveal how much there's random variation at the LSB end of the result.
See the datasheet. FIGURE 21-3: ANALOG INPUT MODEL makes possible to calculate the error limits due the leakage current (quite small) and you can decide how long it takes to charge the sample&hold capacitor to the needed accuracy.
There's suggested adding a capacitor. It reduces the noise but filters also the signal. Simulate it's effect.