Every AI agent you run holds keys that never expire. Issue credentials that do.
Standing keys are unexpiring, over-scoped, and work for any thief. AgentAuth issues expiring, holder-bound passes from a credential office inside your own walls.
A human employee gets a badge, an access review, and an offboarding process; your agents got a password taped under the keyboard.
You locked the building. Then handed out keys that never expire.
You deployed agents because they move faster than people — and every one of them authenticates with a permanent key nobody rotates and nobody watches.
Never expires
Issued last spring, it works today. It will work next year. There is no expiry to reach.
Opens too much
The agent needed one system. The key opens six, because scoping it properly was tedious and nobody did.
Works for any thief
It proves nothing about who holds it. Whoever presents it is the agent — a leaked log line included.
Leaves no record
Used at 3 a.m. from somewhere new, and no record says what it did or on whose behalf.
Agents do not come one at a time — they come in fleets, and every new workflow mints another key.
One leaked machine credential, and the invoices start arriving.
Walk the sequence a single leaked machine credential starts — none of it is hypothetical.
The incident itself
Forensics, containment, and rotation across every system that key could touch — while no record says what it did touch.
The regulatory exposure
If that key could reach customer data, you are in disclosure territory — and nine-figure breach settlements are no longer exotic news.
The board conversation
“We gave software permanent, unaudited access — and we are not sure who else had it” is not a sentence any chief executive wants to own.
The deal that dies in security review
Buyers now ask how your AI agents are credentialed, and “long-lived shared keys” kills contracts you never hear died.
- 01Agent count compounds every sprint — every team that ships an agent ships another credential.
- 02Retrofit cost grows with every agent wired to a standing key. The second-cheapest day to fix this is today.
- 03The question is already on the buyer’s security questionnaire, and on the insurer’s.
The visible cost of a leak is the breach. The invisible cost is every enterprise deal that stalls when your security answers get thin.
Where you have none today, put four locked doors.
Every permanent key, replaced: credentials that expire in minutes, prove their holder, only ever shrink when handed down, and log everything — inside your own walls.
Expires in minutes
Five minutes by default, an hour at the ceiling. A leak becomes a closing window, not an open door.
Proves its holder
Bound to its agent, a pass is useless to a thief. Enforcement is on by default and fail-closed.
Only ever shrinks
A helper’s pass must do less than its parent’s. Escalation is not forbidden by policy; it is impossible by construction.
Logs everything
Every issuance, hand-down, refusal, and revocation lands on a permanent record before the pass is even released.
Asked for more than its parent held — denied automatically, refusal recorded.
On the recordEvery hand-down is capped at eight levels deep, and the full parent-child lineage is kept.
The credential office runs inside your own walls.
The part your security chief asks about first: the issuing office, the records, and the keys all live inside your walls.
Nothing leaves your building
One self-contained installation on infrastructure you already run. If the vendor disappeared tomorrow, it keeps running.
One door, watched
A single published entrance to the whole system. Everything else is unreachable by construction, verified live at every install.
Keys that mind themselves
Signing keys are sealed under a master key only you hold, and replace themselves daily with no outage.
When in doubt, it refuses
Missing a secret, it will not start. Unsure, it denies. There is no skip-security switch in the shipped product.
No entry can be edited. No entry can be deleted.
Source-available under the Apache-2.0 license — your team can read every line before it trusts one.
Three things you would be right to be suspicious of. It is none of them.
Three fair suspicions about any security product — answered plainly.
Not a cloud service
No vendor cloud in the loop. The office, records, and keys are yours, and stay running without us.
Not a migration
Nothing gets ripped out. It sits beside what you run, works with your identity systems; adopt it agent by agent.
Not compliance theater
The controls are live defaults, not checkboxes. Turning security off is an explicit, logged decision.
What runs now — and what you can exercise in your own pilot.
11 lines from the register, every one shipping today and provable on a host you control.
| Ref | Capability | Status |
|---|---|---|
| 01 | Credential issuance — passes minted on demand, living five to sixty minutes; every issuance recorded before the pass is released. | Ships |
| 02 | Holder-bound passes — a stolen pass is useless in anyone else’s hands; enforcement on by default and fail-closed; weakening it is an explicit, logged decision. | Ships · Fail-closed |
| 03 | Shrink-only delegation — hand-downs can only narrow what a pass opens, how long it lives, and where it is accepted; widening refused automatically; depth capped at eight. | Ships |
| 04 | Independent verification and instant revocation — your systems check passes on their own; revocation is immediate, and no hand-me-down pass can outlive the one it came from. | Ships |
| 05 | Self-rotating signing keys — the issuer’s keys replace themselves daily with no outage, and an emergency replacement is immediate. | Ships |
| 06 | Sealed key custody — signing keys sealed at rest under a master key only you hold; the master key itself rotates live; recovery drills documented. | Ships |
| 07 | Tamper-evident record — every action lands in a history the system itself refuses to edit or delete, with full parent-child lineage. | Ships |
| 08 | Administration console and controls — full administrative control of agents, groups, and permissions; the visual console is an optional install. | Ships · Opt-in install |
| 09 | Issuance limits and flood control — per-agent and per-unit limits, with a strict mode that refuses anything unconfigured; the front door slows a flood on its own. | Ships |
| 10 | Works with your identity systems — existing providers plug in; their credentials exchange for narrow internal passes, deny-by-default. | Ships |
| 11 | Backup and point-in-time recovery — two independent restore paths; a restore refuses traffic until it proves it can still sign. | Ships |
Prove it yourself, before you believe a word of this.
Three steps, a finish line agreed in advance, and a decision you make on your own evidence.
Prove it in 30 minutes
One engineer, one afternoon. The full system stands up and runs a scripted self-check that proves the locks are real.
Finish lineEnd to end, on a host you control — without talking to us.
Pilot one workflow
One real agent touching two to five systems, with its standing keys replaced and nothing else disturbed.
Finish lineSuccess is defined up front: zero standing keys left in it, the record answers “who could touch what, when,” and a revocation drill lands under a minute.
Security review in parallel
Your engineers get the technical deep-dive while your security team reviews the assurance package.
Finish lineRecorded adversarial review verdicts, a register of 530 individually verified requirements, and a live self-check run they observe.
AgentAuth is the credential pillar of Iternal’s agentic suite — alongside Ultramemory, the Agent Skills Library, Ultracache, and Iternal Workforce — and each one is adoptable on its own.
The six questions that come up first.
Short answers, no hedging.
Q·01
We already run single sign-on and an identity provider. Why this?
Those systems govern people — browser sign-ins, second factors, and the joiner-mover-leaver process your team spent a decade building. An AI agent is a machine identity none of that machinery covers: no browser, no second factor, no access review. AgentAuth is built for machine principals and runs alongside what you have rather than inside it. The identity systems you already run plug in, and their credentials are exchanged for narrow internal passes.
Q·02
Is this a cloud service we would be wiring our systems into?
No. There is no vendor cloud in the loop. The issuing office, the records, and the signing keys all run on infrastructure you already own, inside your own walls. Nothing phones home, and nothing leaves your perimeter. If Iternal disappeared tomorrow, your installation keeps running exactly as it did the day before. The product is source-available under the Apache-2.0 license, so your engineers can read every line before they trust one.
Q·03
What actually happens when a credential is stolen?
Three things limit the damage. The pass expires on its own within minutes — an hour at the outside — so the window closes without anyone noticing the theft. If the pass was bound to its holder, and that enforcement is on by default, it never worked for the thief at all. Revocation is immediate, and no hand-me-down pass can outlive the one it came from. The record then shows exactly what that pass could reach, so the incident has edges.
Q·04
Can an agent talk itself into more access?
No. When one agent hands work to a helper, the helper’s pass must do less than the parent’s — it can open fewer things, live a shorter time, and be accepted in fewer places. A request to widen any of those is refused automatically, and the refusal goes on the record. Hand-downs are capped at eight levels deep and the full parent-child lineage is kept. This is not a policy someone has to remember to configure; it is a structural property of how passes are issued.
Q·05
What does our auditor get?
A permanent record of every issuance, hand-down, refusal, and revocation, written at the moment the action happens and kept in a history the system itself refuses to edit or delete. Because every hand-down keeps its parent-child lineage, the full chain behind any action can be reconstructed: which agent, acting for whom, with exactly which narrowed permissions, at what time. When the auditor asks who could touch what, and when, you answer from a record rather than a reconstruction.
Q·06
What does it take to try?
One engineer and one afternoon. The full system stands up on a host you control in about thirty minutes and runs a scripted self-check that proves the locks are real, end to end — without talking to us. From there, a two-to-four-week pilot on one real agent workflow: its standing keys replaced, success criteria fixed before you start, and a revocation drill your own team runs. You decide on your own evidence, and nothing is locked in either way.
Question not on this list? Put it on the pilot agenda
The agents are already inside. Make their keys expire.
Weigh what one leaked machine credential costs you — your security chief and your cyber-insurance renewal already priced it — against the enterprise revenue that closes when security review comes back a “yes.”
You will pay one of these two invoices. Only one of them comes with a decision.
Claims on this page trace to the AgentAuth engineering fact base as of July 2026 · Anything labeled illustrative is a concept, not a measurement.