Choose the question that matters
Depth, product judgment, custody, and the willingness to say that an elegant claim is absurd.
About Sekos
Authority belongs outside the model.
Sekos builds infrastructure that lets AI systems propose powerful actions without letting them authorize those actions themselves.
The thesis
A more capable model is not automatically a more authorized one.
Sekos began with a practical control problem. Models can reason, write code, call tools, and operate infrastructure. Their intelligence is useful. Their own assertion that an action is permitted is not.
We place a cryptographically verifiable Gate at the point where a request would cause a consequence. The model can explain, recommend, and ask. An external authority decides whether the exact request may proceed.
The model may propose. It cannot authorize itself.
How the work is done
One builder. One adversarial collaborator.
Formalize and attack it
Rapid construction, theorem work, exhaustive tests, and adversarial review across a large technical surface.
Keep the receipts
Every result carries its scope. Nulls stay visible. A passing demo never becomes a universal theorem by narration.
The builder
Built outside the lab.
Sekos was founded by Ryan McCormick, a 25-year builder and industry veteran of Palo Alto Networks, Amazon, Electronic Arts, and Postman. He left gainful employment in April 2026 to pursue this work full time, partnered deliberately and adversarially with LLMs.
What this is: open research, working software, deployed systems, and falsifiable claims.
What it is not: a promise that a Gate makes an unsafe system safe, or that cryptography can repair an untrustworthy issuer or host.
What it is not: a promise that a Gate makes an unsafe system safe, or that cryptography can repair an untrustworthy issuer or host.
Start with the evidence, then decide whether the architecture fits your boundary.
Read the research