I'm not sure how I feel about this one; we have in the US a right not to incriminate oneself, and it seems like random DNA sampling could very well lead someone to incriminate themselves or a close relative unwittingly. Take this test case, for instance:
A young woman is raped, and her neighbor is suspected based on the victim's physical description. He refuses a DNA test and there isn't a sample on file for him, so they go to his mother and ask her for a sample. Thinking she knows her son to be innocent, since she is also his attested alibi (they had been watching The Price is Right alone together at the time of the alleged crime), she agrees. The results show that she does in fact have very similar DNA to the rapist. Since no one but her can confirm the alibi, but genetic evidence seems very convincing, the jury decides to convict her son of the crime.
Meanwhile, her maternal cousin (the actual rapist, as it turns out, who no one knew was even in the neighborhood at the time) walks free.
My concern here is double: the potential weakness of the method, especially considering just how often crimes are committed by close family, and the likelihood that this sort of genetic entrapment might butt against our constitutional guarantees.
The right not to self-incriminate is to prevent tortured confessions. It normally only applies to what you know, not what you are.
And your scenario would not happen.
Ok, mother takes the DNA test, 50% match, therefore the rapist is an immediate blood relative.
The son's attorney demands a DNA test, it comes back less than 50%. The police are barking up the wrong tree and drop the case.
That would be true, if there was such a thing as a perfect genetic test; But there isn't.
Amplification increases uncertainty, and contamination can never be ruled out. An RFLP test is very good evidence; But it's FAR from perfect. The same is true of fingerprint evidence - it's alleged by prosecutors to be 100% reliable, but a moments consideration would show that this is bullshit - no scientific protocol ever gets 100% accurate results every time, no matter how much effort is put into ensuring experimental conditions that are as ideal as possible; In a criminal investigation, forensic evidence collection is necessarily far from ideal.
Your faith in the science is blinding you to the unavoidability of error in real-world applications.
Real tests don't come back '50% match' (in flashing letters on a console that lights up at a dramatic moment in the investigation); The results that come back say "there is a 99.8% probability of a between 45 and 55% match between these samples, if they are uncontaminated". That an expert witness might well dumb that down to 'it's a 50% match' when testifying in front of a non-expert jury doesn't make it so. That the jurors have probably all seen CSI only adds to the problem.
The scenario suggested by Politesse is FAR closer to reality than your counterexample.
Such testing is of far more value during the investigative phase of a crime, when it can be used to eliminate suspects who do NOT match the profile, than it is during trial, as evidence for the prosecution. The authorities always have an incentive to convict anyone who is charged, and defence attorneys are far from perfect at questioning expert witnesses (after all, they are lawyers, not scientists). The burden of proof is on the prosecutor, and the ability of the prosecutor or expert witnesses to blind both defence and jury with science should be as limited as possible.