If you give a chip more voltage, its transistors will switch faster, but they'll degrade faster. Ideally, you want just barely enough voltage that everything's reliably finished switching and all signals have propagated before it's time for the next clock cycle, as that makes everything work and last as long as possible. When the degradation happens, at first it means things need more voltage to reach the same speed, and then they totally stop working. A little degradation over time is normal, but it's not unreasonable to hope that it'll take ten or twenty years to build up enough that a chip stops working at its default voltage.

The microcode bug they've identified and are fixing applies too much voltage to part of the chip under specific circumstances, so if an individual chip hasn't experienced those circumstances very often, it could well have built up some degradation, but not enough that it's stopped working reliably yet. That could range from having burned through a couple of days of lifetime, which won't get noticed, to having a chip that's in the condition you'd expect it to be in if it was twenty years old, which still could pass tests, but might keel over and die at any moment.

If they're not doing a mass recall, and can't come up with a test that says how affected an individual CPU has been without needing to be so damaged that it's no longer reliable, then they're betting that most people's chips aren't damaged enough to die until the after warranty expires. There's still a big difference between the three years of their warranty and the ten to twenty years that people expect a CPU to function for, and customers whose parts die after thirty-seven months will lose out compared to what they thought they were buying.