AI cracking Bitcoin's cryptography? A Stanford professor and a16z crypto jointly respond, clarifying facts versus speculation and pointing out the crypto opportunities truly worth paying attention to.
Author: a16z crypto
Compiled by: TechFlow
TechFlow introduction: OpenAI's latest mathematical model results have sparked industry panic, with some beginning to worry that AI will destroy the cryptographic systems we rely on to survive. This article invites a Stanford cryptography professor and the a16z crypto research team to clarify what is fact and what is speculation, and to point out the opportunities AI may truly bring. It is worth reading for anyone who depends on cryptographic infrastructure.
After OpenAI's mathematical results prompted warnings from industry insiders and fears that AI could break the cryptography we rely on to survive, we invited cryptographers from Stanford University and experts from a16z crypto to talk about what is fact, what is speculation, what is worth worrying about, and what is not.
Dan Boneh, professor at Stanford University, leads the Applied Cryptography Group and co-leads the Computer Security Lab; he also serves as a special research advisor to a16z crypto
Eddy Lazzarin, general partner at a16z crypto
Justin Thaler, research partner at a16z crypto and associate professor of computer science at Georgetown University
Liz Harkavy, investment partner at a16z crypto
Sonal Chokshi, editor in chief at a16z crypto
In addition, six core conclusions:
1. AI's mathematical progress has not touched the core problems of cryptography
The recently released results demonstrate impressive advances in mathematics and computer science, but they do not include any cryptographic results.
"This is an amazing advance in mathematics and computer science, but there is no cryptography question here at all. ... So you might wonder, does this mean these models are really bad at cryptography? We don't know." Dan Boneh, 03:00
"We did not learn that elliptic curves are weaker than before. We did not learn that any specific cryptographic assumption is less secure than before." Eddy Lazzarin, 15:19
2. Even if AI is smarter, it cannot find an algorithm that does not exist
A large part of modern cryptography is built on the assumption that certain mathematical problems have no efficient solution. This may remain true even for increasingly capable AI.
"We generally believe that these problems do not have efficient algorithms. It is not that we think they might exist and humans are just not smart enough to find them. We think they simply do not exist. If they do not exist, it does not matter how smart AI is. It cannot find that algorithm that does not exist." Justin Thaler, 16:12
"We have to assume that certain problems are hard. We cannot prove that these problems are hard, and the reason is that this relates to a famous unsolved problem in computer science, namely the P versus NP problem. In a sense, we would have to solve that problem first before we could truly argue that certain cryptographic problems are indeed, indeed hard. ... Nothing released this week comes anywhere close to solving P versus NP. We still need to keep making assumptions in cryptography, right?" Dan Boneh, 06:00
3. AI can compress years of cryptographic review time
Trying to break cryptography is itself one of the ways researchers build confidence. A design that withstands sustained, serious attack earns trust. AI may change the speed at which this review happens.
"In the past, when someone proposed an assumption and built a system based on it, we often had to wait around 10 years before deploying the system, in order to build confidence and give people enough time to analyze the assumption. ... And now, suddenly, we can use these models to greatly accelerate that process. ... We can now let a model run for a month trying to break it, and if the model cannot break it, then our confidence that the assumption holds is greatly increased." Dan Boneh, 12:10
4. Quantum readiness provides a head start for some AI risks
Although recent concerns revolve around AI, some of these systems have long been targets of migration efforts because of the threat from quantum computing. This overlap has its limits.
"Because of concerns about the quantum threat, people have already begun to view elliptic curve cryptography as something that may soon become vulnerable. So, for these exact same concerns, we have actually already engaged with, prepared for, and deployed many of the mitigations you would need, except they were originally aimed at a different threat." Eddy Lazzarin, 20:52
"The AI threat is broader in a sense, because it may also actually examine our post-quantum systems and weaken them in some way, or even potentially break them completely." Dan Boneh, 22:08
5. Cryptographic agility allows systems to adapt when weaknesses emerge
Cryptographic agility refers to a system's ability to change its cryptographic methods when evidence or requirements change. It gives engineers the ability to respond to discoveries that cannot be foreseen in advance.
"The systems we deploy can quickly switch from one design to another if the design being used is proven insecure. Cryptographic agility is in a sense our friend in defending against the unexpected." Dan Boneh, 22:47
Dan pointed out that internet connections already use a combination of different cryptographic methods:
"They are actually already using agile cryptography. They use both elliptic curves and lattices for encryption. If you break elliptic curves, that is fine, because it is also encrypted under a lattice-based system." Dan Boneh, 38:40
6. The immediate opportunity: finding more flaws before software ships
For builders, one of the clearest opportunities is to introduce more security testing during development.
Discussing complex cryptographic systems, Justin pointed out:
"The attack surface becomes very large, and yes, I am very optimistic that these AI tools that are making people panic about cryptography will also help us clear out all the vulnerabilities. This is a very hard problem." Justin Thaler, 28:56
"After a system is deployed, it attracts a lot of attention from cryptographers, but by then it is actually too late to attack the cryptographic system. Now, because we control these AI models, any startup deploying new cryptographic capabilities, or even any company deploying any software at all, can have these models analyze vulnerabilities in its cryptography and in its overall software." Dan Boneh, 35:05
"Vulnerabilities can be eliminated. They can be designed around, they can be mitigated, and the main resource we use to mitigate them, AI capability, is getting dramatically cheaper every few months." Eddy Lazzarin, 37:10
Quoted content has been lightly edited for clarity. Ellipses indicate omitted passages.