This is the standard “those ones don’t count” argument.
If someone wants self-identification to be the basis for access to female spaces, they cannot then dismiss every inconvenient male offender as not really trans, or as someone who only identified after conviction. That is just moving the definition when the data becomes uncomfortable.
Also, even if some males do identify after imprisonment, that is not a reason to ignore the risk. It is the point. If policy says a male person’s declared identity can override sex-based boundaries, then institutions have to deal with male offenders who declare that identity. Women and girls are not protected by pretending only the nice, harmless, socially acceptable ones count.
As for the general population, we do not have perfect UK-wide data. But we do have two relevant points.
First, sex offending and serious violence are overwhelmingly male-pattern behaviours. That risk does not disappear because a male person changes name, clothes, pronouns or legal paperwork.
Second, the best-known population register evidence does not show male-to-female transitioners becoming female-pattern in criminality. The Swedish study found they retained a male pattern for crime and violent crime compared with female controls.
Safeguarding is not based on “is this person a real trans person?” It is based on sex, risk, privacy and female people being allowed to say no.
( https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0016885 )