Interesting application of what I might call the "visible effect bias" here. (See: "That Which is Seen and That Which is not Seen"
http://bastiat.org/en/twisatwins.html)
The CPSC sues, complaining that warning labels are ineffective, because parents introduce the toys to their children anyway. The warning labels are ineffective because almost every toy small enough to be put in a mouth has a warning label attached to the effect of "choking hazard; not suitable for children", even if the toys are perfectly safe (which is to say, harm almost never occurs).
Toys are attached with false warning labels because if by some chance a child does choke, either the parents or the CPSC can sue, citing the absence of warning labels. The suitors can win even if the incident was unlikely, due to hindsight bias. Hence, it costs the company nothing to exaggerate the danger by adding a warning label, in order to protect themselves from liability.
So, what is seen by the regulatory system is that putting on a warning label could have prevented such-and-such a case of choking/injury. What is not seen is that adding redundant warning labels decreases the informational value of true warnings, causing increased future cases of choking/injury from truly dangerous products.
Ironically in this case the regulators then sue, arguing that warning labels are insufficient because parents ignore them, due to a problem that was originally created by regulators being too eager to sue!
Now, I haven't looked for any data showing that this is what happened in this case, but it seems to be a common pattern (for example, "may contain traces of peanuts"; and excessive waiting times for FDA approval of livesaving drugs).