[caveat: I am not a Canon expert]
From the photo it appears that the flash is firing at a low power setting, e.g. the reflection over the skinny man with a tie center frame, various pieces of jewelry, etc.
By all appearances the scene appears to have relatively little ambient light and therefore wide aperture, slow shutter and high ISO are implicated as appropriate compensations. On the other hand, a shutter speed that would require high speed sync is contra-indicated and the scene does not suggest motion where first curtain sync would matter.
My understanding is that Evaluative Mode is primarily intended to provide infill light rather than the primary light source. To the degree that's true, the camera aperture, shutter speed, and ISO must be appropriate for shooting with the ambient light.
My understanding of Canon's P-mode is that it allows setting the ISO to a fixed value and constrains shutter speeds to avoid camera shake, i.e. there is a lower bound on shutter speed regardless of available light.
The combination of preset ISO and lens maximum aperture require a shutter speed low enough to create blur due to camera movement. P-mode's "safety" features to prevent blurring from camera shake are raising the shutter speed and over-riding the intent of the flash settings.
Compensating for camera shake has to be done "just in time" or at the last (fraction of a) second so over-riding the flash intent in P-mode makes some reasonable sense given that the alternative is a blurred image. Keep in mind that the point of P-mode and other automatic modes is to reduce the likelihood of very poor images not to produce great ones and to increase the likelihood of capturing acceptable images. Cynically, it's to prevent having to deal with "my camera is broken" complaints from consumers using complex technology.
At a high level, the "what is going on" is that the camera is deciding what is best and because evaluative metering in a low light scene under program mode with a speed light set to high-speed/first-curtain sync is an edge case, it produces its best attempt at an acceptable image. Really knowing what is happening in this circumstance requires understanding implementation details deep down in the software at a level that the programmers themselves may not fully understand.
To me, accurately reasoning out the corner cases of P-mode is harder than reasoning about the scene and setting the camera and speedlight manually because the software can have bugs and light does not. To put it another way, understanding causal relations in P-mode is harder than understanding causal relations with manual settings because it requires understanding vagaries of the program in addition to knowledge of optics and a feel for the light in a scene.