Dan Kaminsky sez,
There's been a lot of talk about some portion of the RSA keys on the Internet being insecure, with "2 out of every 1000 keys being bad". This is incorrect, as the problem is not equally likely to exist in every class of key on the Internet. In fact, the problem seems to only show up on keys that were already insecure to begin with — those that pop errors in browsers for either being unsigned or expired. Such keys are simply not found on any production website on the web, but they are found in high numbers in devices such as firewalls, network gateways, and voice over IP phones.
It's tempting to discount the research entirely. That would be a mistake. Certainly, what we generally refer to as "the web" is unambiguously safe, and no, there's nothing particularly special about RSA that makes it uniquely vulnerable to a faulty random number generator. But it is extraordinarily clear now that a massive number of devices, even those purportedly deployed to make our networks safer, are operating completely without key management. It doesn't matter how good your key is if nobody can recognize it as yours. DNSSEC will do a lot to fix that. It is also clear that random number generation on devices is extremely suspect, and that this generic attack that works across all devices is likely to be followed up by fairly devastating attacks against individual makes and models. This is good and important research, and it should compel us to push for new and interesting mechanisms for better randomness. Hardware random number generators are the gold standard, but perhaps we can exploit the very small differences between clocks in devices and PCs to approximate what they offer.