timer://seal
Your AI acts. Timer Seal answers for it.
Timer Seal sits between an AI and every consequential action it can take. One question — is there authority for this? — three answers: allow, hold for a human, or refuse. Every answer, on the record.
01 The three answers
One question, asked of every consequential action.
Actions covered by a mandate proceed at machine speed, each carrying a seal that traces back to the accountable human who set its bounds.
Decisions that need judgment route to a named person. What the AI proposed, what the human did, and the difference between them — recorded.
Anything no mandate and no human has authorized doesn't happen on the sealed plane. Not filtered as "probably bad" — unauthorized, so undone before it was done.
02 The record
A record no one can quietly rewrite.
Every seal, decision, and refusal is written to an append-only, cryptographically chained ledger, anchored to an independent external witness. Decisions are attributable to named humans. Sensitive content stays off-chain and erasable, so evidence and privacy obligations coexist by design.
We instrument. We never certify. Your auditors draw the conclusions; the ledger holds the facts.
03 Why now
The duties are here. The evidence has to start somewhere.
AI already denies claims, sets prices, sends money, and signs off. Supervisors across Europe now expect documentation, record-keeping, and demonstrable human oversight of exactly these systems — and decisions with legal effect on people carry those duties today, not at some future deadline. Oversight evidence cannot be backfilled: a record that starts now is an asset no late adopter can buy later.
04 The Mirror
Start with the ninety-day Mirror.
We run in pure observation on your real decision traffic. Nothing blocked, nothing disrupted, no workflow change. After ninety days you hold a tamper-evident record of what your AI actually did, who oversaw it, where judgment was real — and which decisions are effectively running on their own today with no named owner.
You keep the record either way. It's yours, in an open format, under your keys.
Request received.
A founder reads every one of these. You'll hear back personally.
We classified ourselves first
Before examining anyone's AI, we ran our own products through our own risk framework and the EU AI Act's high-risk test.
You own the record
Seal works with the AI systems you already run. Your records are exportable, always, at no cost — leaving us never means losing your history.
Research first
Built on the published Human Layer framework. The papers came before the product, and the maturity model is open for anyone to use.