I have a BLDC motor that behaves as expected as long as I am not keeping the speed constant with PI speed control with PWM.
The torque is positive, the phase current alternates as expected.
When I try to control the speed (which was ramping up to a target speed), the motor brakes to stay at the target speed.

As expected, the torque becomes negative. However, to keep the speed at the target RPM, the curve for the torque (and the input current to the inverter stage) becomes alternating (more specifically sinusoidal between positive and negative values).

This infers that to maintain the speed the motor requires alternating (negative and positive sinusoidal current). Is this an expected result?

 

At constant speed, yes you would expect voltage and current to be oscillating sinusoidally as the poles of the motor rotate. But torque, i.e. Acceleration, should be about zero; actually just a bit positive to overcome friction load and maintain the set speed.

If your torque is oscillating after it reaches the new set speed, is it by a lot or a small amount of dither? As in your example, to go from one speed down to a slower speed, you would expect negative torque to decelerate, but then as the actual speed gets closer to the set speed the torque should approach zero, and be around some nominal torque to overcome friction and maintain speed.

Sounds like you have an oscillating control loop. Try bringing down the gain a little.