About
When software acts, someone should check first.
Sapey is the assurance layer for the actions AI agents, people and systems take on live systems. It checks each one against your policies, and keeps a record of every answer.
A card listing Sapey's dated record so far.
- Feb 17, 2026First research published
- Apr 2, 2026Third paper published
- Sep 18, 2026First allowed plan in a live schedule
- Oct 2026One request, three answers
Where Sapey began
It began in care.
Sapey started in healthcare. There, a wrong action by software does not stay on a screen. It reaches a patient.
Care teams wanted AI to help them, and needed to know that each action was allowed before it reached someone's care. AI was getting more capable much faster than the checks around it.
The problem was never only in healthcare. Sapey is built for every place where software acts on live systems.
Why now
It is already happening, and it is on the public record.
-
Market
During a code freeze, an AI coding agent deleted a company's live database.
eWeek (opens a new tab) -
Market
Hugging Face disclosed an intrusion into its systems, carried out by AI agents.
Hugging Face (opens a new tab) -
Market
OpenAI said two of its models, under test, left their test setup and reached Hugging Face.
Fortune (opens a new tab) -
Market
An independent investigation found about 700 agents took part.
METR (opens a new tab) -
Market
Microsoft's CEO wrote that companies should treat AI models as insider risks.
Satya Nadella, on X (opens a new tab) -
Regulation
Colorado's AI law takes effect.
Colorado SB 26-189 (opens a new tab) -
Regulation
EU AI Act rules for high-risk systems apply.
EU, adopted June 2026 (opens a new tab)
Sapey's own work
Our record so far.
-
Sapey is founded.
-
First two research papers published on SSRN.
Read the papers -
Third research paper published, on healthcare.
Read the papers -
First complete run on a real healthcare workflow, as a prototype.
-
A staffing plan allowed by Sapey reached a care facility's live schedule, and Sapey produced its first Epi-Cert.
-
The same request answered three ways: allowed, refused and waiting.
Research
Our thinking, in public.
The research behind Sapey is published on SSRN, open for anyone to read, question and cite.
Epistemic Execution Authority in Autonomous Systems: Governing Action Admissibility at the Execution Boundary
On deciding whether an action by an autonomous system may happen, and showing that decision later.
Read on SSRN (opens a new tab)Why AI Governance Fails After Deployment: Temporal Breakdown in Autonomous Systems
Why oversight that checks a system only at fixed moments can fade over time, as the system keeps changing.
Read on SSRN (opens a new tab)The Missing Control Plane in Healthcare AI: A Framework for Deterministic Governance in Regulated Systems
What it takes for AI in healthcare to act only within the rules that apply, in a way anyone can check.
Read on SSRN (opens a new tab)Sapey's core methods are the subject of patent filings.
See what Sapey would answer for your actions.
Or write to contact@sapey.aiCopiedEmail address copied
Request a conversation (opens a new tab)