CharlieKey takeaways from your favorite podcasts
← 14 October 2026a16z crypto

Can AI Actually Break Cryptography? Experts Weigh In On the FUD

AI's math progress does not change today's cryptographic security; labs should disclose their attempts to break standard assumptions, and existing backups plus crypto agility mean there's no reason to panic.

Watch on YouTube ↗

Charlie's summary

  1. Don't panic: nothing about cryptographic security has changed, and the OpenAI math results contained no cryptography at all

    Dan stresses that the world is the same as a year ago. Of the roughly 700 papers OpenAI posted, none addressed cryptography, even theoretical work. Dan sees this absence as conspicuous and calls on the frontier labs to say whether they tried attacking standardized crypto. Until something new is learned about our standard systems, Dan says, keep using them as they are.

    ▶ 1:29
  2. Cryptography rests on unproven hardness assumptions, so failed attacks, including AI's, are what build confidence

    Dan explains that no one can prove problems like elliptic curve discrete log or lattice problems are hard without resolving P versus NP. Confidence comes from decades of people trying and failing to break them. A clever human could have broken an assumption 10 years ago, so AI is not a new kind of risk. If the labs fail, or only shave off “10 bits of security”, that is strong evidence the assumptions are worth using.

    ▶ 5:36
  3. Labs can compress a decade of cryptanalysis into one expensive run, giving more confidence than ever

    Normally a scheme waits about 10 years of expert scrutiny before it earns trust. Eddie imagines a lab spending “$20 million, $50 million” with its most capable unrestricted model trying to break a scheme. A failure would build more conviction than the conventional process gives in a decade. Dan adds that anyone designing a new assumption could let a model attack it for a month. The gap is that the labs aren't sharing what they find.

    ▶ 11:10
  4. An AI-found break would be catastrophic, but backups exist: NIST is standardizing other public key systems

    Dan says a practical break of elliptic curve discrete log would hit the whole digital world, from banking to healthcare. But NIST is already standardizing many other public key systems, so there are fallbacks. Dan lists isogenies, code-based, multivariate and hash-based systems. The missing piece is knowing how urgent the move is, and that lack of information is what's creating the panic.

    ▶ 9:38

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.