You may have seen the preprint posted today by Lenstra et al. about entropy problems in public keys. Zakir Durumeric, Eric Wustrow, Alex Halderman, and I have been waiting to talk about some similar results. We will be publishing a full paper after the relevant manufacturers have been notified. Meanwhile, we’d like to give a more complete explanation of what’s really going on.
We have been able to remotely compromise about 0.4% of all the public keys used for SSL web site security. The keys we were able to compromise were generated incorrectly–using predictable “random” numbers that were sometimes repeated. There were two kinds of problems: keys that were generated with predictable randomness, and a subset of these, where the lack of randomness allows a remote attacker to efficiently factor the public key and obtain the private key. With the private key, an attacker can impersonate a web site or possibly decrypt encrypted traffic to that web site. We’ve developed a tool that can factor these keys and give us the private keys to all the hosts vulnerable to this attack on the Internet in only a few hours.
However, there’s no need to panic as this problem mainly affects various kinds of embedded devices such as routers and VPN devices, not full-blown web servers. (It’s certainly not, as suggested in the New York Times, any reason to have diminished confidence in the security of web-based commerce.) Unfortunately, we’ve found vulnerable devices from nearly every major manufacturer and we suspect that more than 200,000 devices, representing 4.1% of the SSL keys in our dataset, were generated with poor entropy. Any weak keys found to be generated by a device suggests that the entire class of devices may be vulnerable upon further analysis.