Proven foundations
Constructions rest on NIST-approved families and published reductions. Optimization never invents a primitive.
Technology
How one works
ONE makes the handshake programmable. Identity, policy and encryption become one trust object. The object rides with the payload and is verified at every hop.
Trust is a sequence, not a feature
- 1
handshake with identity
- 2
secure channel established
- 3
trust object forged
- 4
verified at every hop
step 1 of 4
Who is connecting is a field of the connection.
identity + policy + encryption = one object, riding with the payload
Handshake
Ten fields. Zero extra round trips.
A ten-megabyte payload arrives over TLS. The receiver learns its size and nothing else. With ONE, the handshake carries identity, purpose, policy, device, context, risk, lifecycle, compliance and score.
Length unlocked
How much is moving. TLS already carries this. ONE keeps it, then adds nine more.
Unlocking it: The connection knows its size before the first byte. Overflow and padding have nowhere to hide.
Select a field to see what unlocking it does.
The missing derived function
A cipher that expands. A handshake that is the tweak.
Length is the only field TLS carries for free. Everything else — who, why, for how long, on which device — lives in software, where it can be ignored. AES still consumes 128-bit blocks. NIST opened Accordion (IR 8552 / SP 800-197A) because a variable-input-length cipher is still an unsolved suite problem. The derived function this industry does not ship: the handshake program is the tweak on that object. Decrypt only in-terms. Ciphertext anywhere else.
Open problem
Variable-length object. Handshake as the tweak.
NIST opened Accordion (IR 8552) because AES still eats 128-bit blocks. TLS still carries length and nothing else. Policy still lives in the application, where it can be stripped. The missing derived function: the ten fields of the handshake become the tweak on a variable-length object.
Proof
Provably secure. Quantitatively stated. Combinatorially hard.
Post-quantum primitives rest on proven foundations. Every signed version ships a complexity statement. The threat model is a no-intervention machine running on a quantum device, fully autonomous. The cipher is built so that machine still faces a combinatorial search.
threat model: a no-intervention machine on a quantum device, fully autonomous
no human in the loopAlgebraic (collapses)
RSA-2048 · ECC P-256 · Shor → polynomial
Combinatorial (holds)
ONE KEM · object cipher · combinatorial search
| Construction | Model | Bound shipped |
|---|---|---|
| RSA-2048 | Algebraic | polynomial (Shor) |
| ECC P-256 | Algebraic | polynomial (Shor) |
| ML-KEM | Module-LWE | published reduction |
| ONE KEM | LWE | reduction under NDA |
| Object cipher | LWE-family reduction | combinatorial search |
Quantitative complexity
Every signed version ships a complexity statement: classical bits, quantum bits, and the bound the proof actually gives, not a slogan.
Autonomous quantum adversary
The threat model is a no-intervention machine running on a quantum device. Fully autonomous. No human in the loop.
Combinatorial hardness
ONE ciphers are built so that even that machine faces a search space that stays combinatorially complex. Algebraic structure is not the load-bearing wall.
Language
You write the rule in plain language.
Instead of configuring a dashboard or writing code, the owner states the rule in a sentence and the system compiles it into the connection.
The rule, as someone states it
no model may train on this corpus; the finance team in the EU may read it for 9 days; every read is provable
What the system compiles it into
train = denymodel = nonegroup = financegeo = EUpurpose = readterm = 9dwitness = req… n
Each clause becomes a condition on the connection. Anything that fails a clause never receives a key.
What it binds
ONE makes the handshake programmable.
Identity
Who or what is connecting. Verified inside the connection, not at a door the traffic can route around.
Purpose and behavior
What the person or agent is allowed to do. A stolen credential used outside that purpose is cut off.
Policy
Rules written into the agent. Run for this period then expire. Do not exceed this cost. Never replicate.
Device
Operate only on this device or group, never leave it, and connect only where the same rules are present.
AI context
The boundaries the agent works within, set by you rather than guessed by the model.
Lifecycle
Created for one purpose, limited in time and scope, and shut off on schedule instead of running forever.
Where encryption can live
At rest and in transit: solved. Everywhere else, data is exposed.
Data moves through ten states. Encryption reaches two of them. The other eight are where an AI agent does its work. ONE enforces from inside the data: one layer, every state, at the speed of use.
StateTodayWith ONE
At rest
files and disks
In transit
the pipe between machines
In use
while the work happens
In a shared cache
reused state that must not cross tenants
In an agent
identity, authority, actions
In a model’s authority
what it may read, until when
In a live data flow
packets with an owner and a purpose
In collaboration
shared without surrendering control
Across clouds and borders
residency enforced by the key
In a delegation
authority that can only narrow
Cryptography
We didn’t replace cryptography. We extended it.
Traditional encryption protects data. ONE also protects identity, policy, and intent, with ONE constructions: a context-aware KEM and a variable-size object semantic cipher. Specifications and security reductions available under NDA.
- ONE constructions, not a wrapper. Reductions under NDA
- ONE KEM: LWE, parameters to the bound, keys about half at matched security
- Variable-size object semantic cipher. Terms inside the object
- Programmable handshake that carries ten fields, not one
Industry PQONE KEM
Compute
ML-KEM, matched security
at or below baseline
Key size
ML-KEM, matched security
about half
Block
fixed
variable, larger
Context
length only
ten fields in the handshake
Reduction
published (Module-LWE)
available under NDA
Every construction rests on an assumption. ML-KEM’s reduction to Module-LWE is published. ONE’s reduction sits on LWE and is available under NDA. Public figures are at matched security, not 512 versus 768. methodology.
Agility
Existing primitives. Hybrids. ONE inventions.
The SDK is how a client takes this. Pick from the factory. Compose a hybrid. Keep what already runs. Add the programmable handshake only where it must.
Classical
- AES-256-GCM
- ChaCha20-Poly1305
- X25519
- ECDSA
Post-quantum
- ML-KEM
- ML-DSA
- SLH-DSA
Constrained
- Ascon
- SPHINCS+ tiny
- symmetric-first
ONE
- Context-aware KEM
- Object semantic cipher
- Programmable handshake
Hybrids
- ML-KEM ∥ X25519
- AES-GCM ∥ ONE AEAD
- classical + PQ dual
Validation
You see it as it happens. You stop it as it happens.
Continuous validation
Verified at every hop, not only at the door.
Immutable audit
A permanent record sealed into the payload as it moves.
Stop as it happens
The same layer that reports is the layer that refuses.
Technology
AI can be governed.
The agent is stopped before it acts, every interaction is visible as it happens, the evidence writes itself, and the AI runs no slower for any of it.