Can AI really break cryptography? Experts weigh in
Short summary
Cryptographers from Stanford and a16z address whether recent AI breakthroughs threaten the math underpinning encryption. The short answer: recent AI progress on general math problems tells us almost nothing about cryptographic security, and even smarter AI can't solve problems that don't have efficient solutions.
- Recent AI math breakthroughs didn't include any cryptography results, so they reveal nothing about whether encryption methods are actually vulnerable.
- Modern cryptography assumes certain math problems have no efficient algorithm at all—not just that humans haven't found one yet. A smarter AI can't discover an algorithm that doesn't exist.
- AI could compress the years-long vetting process for new cryptographic systems by rapidly testing whether they can be broken, building confidence faster than traditional analysis.
- Cryptographic agility—the ability to switch between different encryption methods—already exists in systems like TLS and offers defense against both AI and quantum threats.
Full summary
Recent AI breakthroughs in mathematics sparked worry that artificial intelligence could crack the cryptographic systems protecting everything from banking to crypto wallets. The reality is more nuanced. OpenAI's results showed genuine progress on hard math problems, but they didn't actually touch cryptography itself. The key insight from cryptographers like Stanford's Dan Boneh is that modern encryption relies on the assumption that certain mathematical problems have *no* efficient solution—not that we haven't found one yet, but that one doesn't exist. No matter how smart AI becomes, it can't find an algorithm that isn't there. We can't mathematically prove these problems are truly unsolvable (that's tied to the P versus NP problem, one of computer science's biggest open questions), so cryptography will always rest on assumptions. But that's how it's always worked.
The practical upside is that AI could actually strengthen our confidence in cryptography faster. Historically, new cryptographic systems needed roughly a decade of scrutiny from researchers trying to break them before deployment. AI could compress that timeline dramatically—run an AI model for a month attempting to crack something, and if it fails, we gain tremendous confidence in that system's security. There's also existing overlap with quantum computing defenses. Because the industry has already begun shifting away from elliptic curve cryptography due to quantum threats, many mitigation strategies are already in place. Some systems now layer multiple encryption methods simultaneously, so breaking one doesn't matter if another still holds.
For builders and companies deploying new software, the clearest opportunity is using AI for security testing *before* systems ship, not after. Cryptographic systems have enormous surface area for bugs, and AI tools can catch vulnerabilities during development when they're far cheaper to fix. As one investor noted, the primary resource for mitigating vulnerabilities—AI capabilities—gets radically cheaper every couple months. The real risk isn't that AI breaks cryptography tomorrow; it's deploying insecure systems today. Cryptographic agility (designing systems that can switch to different encryption methods if needed) also gives engineers a safety valve for unexpected discoveries down the road.