SHA-1 Freestart Collision

There's a new cryptanalysis result against the hash function SHA-1:

Abstract: We present in this article a freestart collision example for SHA-1, i.e., a collision for its internal compression function. This is the first practical break of the full SHA-1, reaching all 80 out of 80 steps, while only 10 days of computation on a 64 GPU cluster were necessary to perform the attack. This work builds on a continuous series of cryptanalytic advancements on SHA-1 since the theoretical collision attack breakthrough in 2005. In particular, we extend the recent freestart collision work on reduced-round SHA-1 from CRYPTO 2015 that leverages the computational power of graphic cards and adapt it to allow the use of boomerang speed-up techniques. We also leverage the cryptanalytic techniques by Stevens from EUROCRYPT 2013 to obtain optimal attack conditions, which required further refinements for this work. Freestart collisions, like the one presented here, do not directly imply a collision for SHA-1.

However, this work is an important milestone towards an actual SHA-1 collision and it further shows how graphics cards can be used very efficiently for these kind of attacks. Based on the state-of-the-art collision attack on SHA-1 by Stevens from EUROCRYPT 2013, we are able to present new projections on the computational/financial cost required by a SHA-1 collision computation. These projections are significantly lower than previously anticipated by the industry, due to the use of the more cost efficient graphics cards compared to regular CPUs. We therefore recommend the industry, in particular Internet browser vendors and Certification Authorities, to retract SHA-1 soon. We hope the industry has learned from the events surrounding the cryptanalytic breaks of MD5 and will retract SHA-1 before example signature forgeries appear in the near future. With our new cost projections in mind, we strongly and urgently recommend against a recent proposal to extend the issuance of SHA-1 certificates by a year in the CAB/forum (the vote closes on October 16 2015 after a discussion period ending on October 9).

Especially note this bit: "Freestart collisions, like the one presented here, do not directly imply a collision for SHA-1. However, this work is an important milestone towards an actual SHA-1 collision and it further shows how graphics cards can be used very efficiently for these kind of attacks." In other words: don't panic, but prepare for a future panic.

This is not that unexpected. We've long known that SHA-1 is broken, at least theoretically. All the major browsers are planning to stop accepting SHA-1 signatures by 2017. Microsoft is retiring it on that same schedule. What's news is that our previous estimates may be too conservative.

There's a saying inside the NSA: "Attacks always get better; they never get worse." This is obviously true, but it's worth explaining why. Attacks get better for three reasons. One, Moore's Law means that computers are always getting faster, which means that any cryptanalytic attack gets faster. Two, we're forever making tweaks in existing attacks, which make them faster. (Note above: "...due to the use of the more cost efficient graphics cards compared to regular CPUs.") And three, we regularly invent new cryptanalytic attacks. The first of those is generally predictable, the second is somewhat predictable, and the third is not at all predictable.

Way back in 2004, I wrote: "It's time for us all to migrate away from SHA-1." Since then, we have developed an excellent replacement: SHA-3 has been agreed on since 2012, and just became a standard.

This new result is important right now:

Thursday's research showing SHA1 is weaker than previously thought comes as browser developers and certificate authorities are considering a proposal that would extend the permitted issuance of the SHA1-based HTTPS certificates by 12 months, that is through the end of 2016 rather than no later than January of that year. The proposal argued that some large organizations currently find it hard to move to a more secure hashing algorithm for their digital certificates and need the additional year to make the transition.

As the papers' author