Here is what happened, in one paragraph: a Johns Hopkins cryptography professor warned this month that AI could undermine public-key cryptography — the math the internet's trust runs on. Nothing has been broken, and he said so himself. But the warning exposes a structural fact: most systems bet everything on one kind of lock. We didn't. This post covers the warning, with its receipts — then the fix we invented for it: Scripture Armor, a transport design whose secrecy layer never rested on public-key math in the first place, wrapped in an armor of ancient-looking glyphs that were never a language.
What happened?
On October 8, 2026, Matthew Green — a cryptography professor at Johns Hopkins University — posted: "I think we might lose public key cryptography" (reported by BeInCrypto, October 2026). The post followed OpenAI's October 6, 2026 announcement that an unreleased AI model had produced 722 mathematical manuscripts while attempting roughly 4,000 unsolved problems.
Has anything actually been broken?
No. The 722 manuscripts are still being evaluated, and coverage of the release states plainly that nothing in them shows a way to defeat the encryption in use today (Golden Era, reporting on Fortune and The Independent's assessment, October 2026). Green clarified that he knows of no imminent breakthrough — his concern is that AI could find mathematical weaknesses decades of human research missed, and he cautioned against panic (Wordfence, October 2026). He also made the point this post is built on: public-key schemes operate on thin margins, while symmetric cryptography carries far more safety margin (WebProNews, October 2026).
Why is it a "one lock" problem?
Because public-key cryptography is a bet on specific math problems staying hard. RSA rests on the difficulty of factoring very large numbers; elliptic-curve systems rest on a related problem. History says those bets expire on their own schedule: in 1977, factoring the famous RSA-129 number was estimated to take 40 quadrillion years — it was factored in 1994, seventeen years later. No quantum computer required; better mathematics and more machines were enough. If AI accelerates mathematics itself, every timeline built on "this problem is too hard" becomes a guess, not a guarantee. And the exposure has already started: under "harvest now, decrypt later," adversaries record encrypted traffic today and store it until a breakthrough makes it readable (Sillectus post-quantum cryptography guide, October 2026). Data that crossed a wire under a breakable scheme is already on a clock.
Don't the post-quantum standards already fix this?
They fix the public-key half, and the work matters. NIST finalized three post-quantum standards on August 13, 2024: FIPS 203 (ML-KEM) for key establishment, FIPS 204 (ML-DSA) for digital signatures, and FIPS 205 (SLH-DSA) for hash-based signatures. Hybrid key exchange — classical X25519 combined with ML-KEM — is already shipping in Chrome, Cloudflare, Signal, and Apple's platforms, and NIST proposes deprecating RSA and elliptic-curve systems after 2030 and disallowing them after 2035 (Sillectus, October 2026). Migrate — but notice what migration is: a swap of one set of mathematical assumptions for newer ones. Necessary. Not, by itself, an architecture.
The fix I invented: Scripture Armor
Scripture Armor is the transport design we're building into Kingmaker, our in-development local AI system at Ronin Inc in Phoenix, Arizona. It is a design, not a shipped product — we label it that way everywhere, including here. Its premise is the one Green's own footnote points at: don't rest the secrecy on the lock that's being circled. Three layers:
The secrecy never depended on the lock
The layer that actually keeps secrets is symmetric AEAD encryption — AES-256 and ChaCha20-Poly1305. Symmetric ciphers do not rest on factoring or discrete logarithms, so the attacks that threaten RSA and elliptic-curve systems — classical, AI-assisted, or quantum — do not apply to them the same way; quantum search only halves their effective strength, leaving a 256-bit key with roughly 128 bits of post-quantum margin. If public-key cryptography fell tomorrow, this layer's math would not notice. There is no lock here to pick.
The visible armor: an ancient language that was never a language
Over the ciphertext rides the part you can see — and the part designed to waste an attacker's century. Every packet is re-encoded in a glyph armor drawn from undeciphered-script-class signs (Indus-class glyphs, carried as Private Use Area codepoints that don't even render without our bundled font, with encoded scripts mixed in as decoys). This is not a static substitution cipher: the symbol alphabet's permutation is re-derived for every packet from that packet's nonce, so the encoding itself is keyed and changes packet to packet. What an inspector sees is an ancient text. What it actually is: ciphertext, wearing a script that was never a language, in an alphabet that won't sit still.
A signed frame that testifies
Every packet travels inside a signed metadata frame — signature, canary ID, expiry. Tamper with a packet inside our fleet and it is rejected and quarantined on the spot. Material stolen from the fleet carries its canary with it; when it wakes up somewhere it shouldn't be, it phones home. Visibility, not counterstrike, is the weapon — sight over strike. An optional public blacklist (off by default, published only on the owner's tap) turns the receipts into a public record: attacking IPs, timestamps, behavior class, and the signed evidence hash. The evidence does the talking.
The honest part: what is swappable
We will not claim the whole design is public-key-free, because it isn't — and pretending otherwise is how the industry got here. The metadata frame's signature is classical public-key cryptography (Ed25519). So it is treated the way every load-bearing assumption should be: as the designated swappable part. Our release keys rotate with every release, and the launcher verifies against an append-never-replace bundle of keys, so retiring a key or an algorithm never orphans a legitimate install (public record: ronin-launcher, commit 40ae309). The honest claim is strong enough without decoration: the secrecy never depended on the lock they're about to pick — and every part that did is built to be swapped.
FAQ
Did AI break encryption in October 2026?
No. An unreleased OpenAI model produced 722 mathematical manuscripts that are still unverified, and reporting on the release states that nothing in them shows a way to break today's encryption (Golden Era / Fortune, October 2026). Matthew Green's warning concerns what AI-driven mathematics could find next, not a break that has happened.
What did Matthew Green mean by "lose public key cryptography"?
Green, a Johns Hopkins cryptography professor, posted the warning on October 8, 2026 (BeInCrypto, October 2026) and clarified that he knows of no imminent breakthrough. His concern is that AI could discover weaknesses in the mathematical assumptions under RSA and elliptic-curve systems that human researchers have missed (Wordfence, October 2026).
What is Scripture Armor?
Scripture Armor is Ronin Inc's transport design, in development as part of Kingmaker. Its secrecy layer is symmetric AEAD encryption (AES-256 / ChaCha20-Poly1305), which never depended on factoring; over the ciphertext rides a Rosetta-keyed glyph armor — an undeciphered-script-class symbol alphabet whose permutation is re-derived per packet; and every packet carries a signed metadata frame with a canary ID, so tampering is detected and stolen material reports where it wakes up. It is a design in development, not a shipped product.
Is Scripture Armor public-key-free?
No — and we say so on purpose. Its packet signatures are classical public-key cryptography, deliberately isolated as the designated swappable part under our crypto-agility doctrine (per-release key rotation, append-never-replace verification). The secrecy layer is the part that never used the lock.
What is harvest now, decrypt later?
It's the practice of recording encrypted traffic today in order to decrypt it whenever a future breakthrough makes that possible (Sillectus, October 2026). It means traffic transmitted under breakable schemes is exposed retroactively — and why what never leaves your own machine can't be harvested at all.
Are the post-quantum standards ready to use?
Yes. NIST finalized FIPS 203 (ML-KEM), FIPS 204 (ML-DSA), and FIPS 205 (SLH-DSA) on August 13, 2024; hybrid post-quantum key exchange is already deployed by Chrome, Cloudflare, Signal, and Apple; and NIST proposes deprecating RSA and ECC after 2030 (Sillectus, October 2026). We treat them as the migration path for the parts that do use public-key math — not as a substitute for architecture.
Why does symmetric encryption hold up better?
Symmetric ciphers like AES-256 and ChaCha20 don't depend on factoring or discrete logarithms, so the attacks that threaten RSA and elliptic-curve systems don't apply to them; quantum search only halves their effective strength, leaving a 256-bit key with roughly 128 bits of post-quantum margin. Green himself noted symmetric schemes carry far more safety margin (WebProNews, October 2026).
Sources: BeInCrypto (Oct 2026); Wordfence (Oct 2026); WebProNews (Oct 2026); Golden Era, summarizing Fortune / The Independent (Oct 2026); Sillectus, "Harvest Now Decrypt Later and PQC Explained" (Oct 2026); NIST FIPS 203/204/205 (Aug 13, 2024); Scripture Armor design spec, Ronin Inc (in development — Kingmaker, to be released); ronin-launcher public commit history (commit 40ae309).
