Prying open Pandora's box: KCI attacks against TLS

Protection of Internet communication is becoming more common in many products, as the demand for privacy in an age of state-level adversaries and crime syndicates is steadily increasing. The industry standard for doing this is TLS. The TLS protocol supports a multitude of key agreement and authentication options which provide various different security guarantees. Recent attacks showed that this plethora of cryptographic options in TLS (including long forgotten government backdoors, which have been cunningly inserted via export restriction laws) is a Pandora's box, waiting to be pried open by heinous computer whizzes. Novel attacks lay hidden in plain sight. Parts of TLS are so old that their foul smell of rot cannot be easily distinguished from the flowery smell of 'strong' cryptography and water-tight security mechanisms. With an arcane (but well-known among some theoretical cryptographers) tool, we put new cracks into Pandora's box, achieving a full break of TLS security. This time, the tool of choice is KCI, or Key Compromise Impersonation.

The TLS protocol includes a class of key agreement and authentication methods that are vulnerable to KCI attacks: non-ephemeral Diffie-Hellman key exchange with fixed Diffie-Hellman client authentication - both on elliptic curve groups, as well as on classical integer groups modulo a prime. We show that TLS clients that support these weak handshakes pose serious security concerns in modern systems, opening the supposedly securely encrypted communication to full-blown Man-in-the-Middle (MitM) attacks.

This paper discusses and analyzes KCI attacks in regard to the TLS protocol. We present an evaluation of the TLS software landscape regarding this threat, including a successful MitM attack against the Safari Web Browser on Mac OS X. We conclude that the insecure TLS options that enable KCI attacks should be immediately disabled in TLS clients and removed from future versions and implementations of the protocol: their utility is extremely limited, their raison d'être is practically nil, and the existence of these insecure key agreement options only adds to the arsenal of attack vectors against cryptographically secured communication on the Internet.