That is true. Utility would not be concave anymore under prospect theory (only for gains), but convex for losses, which is evidence against CAPM.
CAPM is valid either :
-if the utility function is quadratic (which is nonsense in terms of economic interpretation, and in general, Von Neumann- Morgenstern utility describes poorly reality and should be rejected as a descriptive Theory)
-if the distribution of the returns is elliptic, that is a general class that includes the normal distribution (which is nonsense also because there is skewness in real life).
In fact, CAPM is never verified, and fails miserably empirically.
But even Prospect Theory cannot fix the problem at an equilibrium/aggregate level, since it doesn't solve the Pricing Kernel Puzzle. That is because Prospect Theory doesn't aim to solve this problem, but rather describes individual behavior in certain circumstances (simple lotteries).
If you want to really describe what happens, the utility has to alternate concavity and convexity, both in the domain of gains and losses (double alternation, with a more pronounced concavity for losses and convexity for gains, the exact opposite of Prospect Theory), while being globally concave and keeping the framing effect which is still very important.
In fact, we can derive what I call Real Utility functions (that does not depend of preferences nor risk aversion but rather imply both of them), that arises from overconfidence, underestimation of extrema and non bayesian learning. But the approximation of real utility functions is what is called the CAPM^2 (for continuous asymetric polynomial models).That was my master thesis actually,so I know it well.
How to integrate that utility into a Beta model requires 3 pages of calculus, but it is doable and has been done. The beauty is that it works smoothly and beats off Fama
French factors plus momentum.