Clarification about Phong lighting model

Question

I've read some tutorials about Phong and Blinn lighting models and tried to implement a rendered myself. To my knowledge, the basic approach consists of the following steps:

• calculate the vector from the current fragment to the light rayToLight (and normalize it)
• calculate the diffuse color intensity using diffuseIntensity = max(dot(rayToLight, normal))
• calculate the vector from the current fragment and the camera rayToCamera (and normalize it)
• reflect rayToLight: reflected = reflect(-rayToLight, normal)
• calculate the specularIntensity = pow(max(dot(reflected, rayToCamera), 0.0), shininess).

The problem is that specularIntensity ( dot(reflected, rayToCamera) to be precise), can be non-zero even if the light and the camera are on the opposite sides of a surface. Take the following example It is possible that the angle between reflected and rayToCamera is less than 90 deg even if the camera and the light are on the opposite sides of a surface. This means that despite the surface is not lit by the diffuse light, it can show specular reflections which is, of course, not wanted. I wasn't aware of this problem until it occurred when I was trying to implement a simple phong lighting model.

Have I done something wrong or is it just the way it should work? Thanks