Governance
elan1 Govern: Agent governance and assurance
Decides before the agent acts.
Your rules run before any agent action does. Allow, ask a person, or refuse, and every outcome ends in a record your auditor can check. Every action, answerable.
What is AI agent governance?
AI agent governance is the set of controls that decide, before an AI agent acts, whether the action runs, waits for a person's approval or is refused, and that record the outcome. elan1 Govern checks every proposed action against your rules, sends risky ones to a named approver and writes a signed record your auditor can verify.
The problem
Most AI tools watch. Few stop anything.
Enterprises have bought tools that observe AI after the fact. But the moment of consequence is the action: a payment released, a credit limit changed, a contract amended. Almost nothing stands in front of it.
60%
of enterprises cannot stop an AI agent that misbehaves, and 65% have already had an incident caused by one.
Market: CSA/Token, April 2026
How it works
Ask, check, approve, act, record. Every time.
Step 01
Ask
The agent proposes an action instead of taking it.
Step 02
Check
Govern confirms who the agent is and reads your rules.
Step 03
Approve
Routine actions pass. Risky ones wait for a named person. The rest are refused.
Step 04
Act
Only what was decided runs, with credentials the agent never sees.
Step 05
Record
Every outcome is written to a record your auditor can verify.
Ask
The agent proposes
Check
elan1 Govern checks
- Who the agent is
- Your rules, read first
Approve
Three outcomes
- AllowRoutine actions pass, and run with credentials the agent never sees.
- Ask a personRisky actions wait for the person you name. Unanswered after 24 hours, they are refused.
- RefuseAnything not clearly allowed. An error refuses too.
Act and record
The record
Always available
Emergency stop
What it does
Your rules before the action. Proof after.
01
Decides before it runs
Every agent action is checked against your rules first. Anything not clearly allowed is refused, and an error refuses too.02
Asks a named person
Risky actions, such as large payments, bank-detail changes and deletions, wait for the person you name. They answer in the console, or with one tap in Slack or WhatsApp.03
Records every outcome
Allowed, approved or refused, every decision leaves a signed record. Auditors check it offline, without trusting elan1.04
Stops everything
One emergency stop pauses every agent at once and starts the incident report.
The stage of each capability is in our claims register.
The record
A record a stranger can check.
Allowed, approved or refused, every decision leaves a signed record. Your auditor can verify an elan1 Govern record offline, with one command, without trusting elan1.
The record is yours. You can export it at any time, and it stays yours if you leave.
16
fields in every action record.
Product fact
One action, on the record
elan1 Govern- Proposed
- What the agent asked to do
- Agent
- Who the agent is, as confirmed by Govern
- Rules
- Which of your rules applied
- Outcome
- Allowed, approved or refused
- Decided by
- The rule, or the person you named
- Result
- What ran, if anything
Assurance levels
How far you can check each claim.
Every elan1 Govern claim carries one of three levels: how far you can check it without trusting us. This is where each one stands today.
Level 1 · Stated
We state it.
The claim is ours, written down and signed off. You check it against our documents and contracts.
- GOV-06One emergency stop pauses every agent at once and starts the incident report.
- GOV-07A refusal costs 40% of one governed action.
- GOV-08Gmail and Zoho Books connectors are built in.
Level 2 · Verifiable
You can check it yourself.
The claim leaves evidence in the record, and your auditor can test it offline, without asking us.
- GOV-01Every agent action gets one of 3 outcomes: allow, ask a person, refuse.
- GOV-02Anything not clearly allowed is refused, and an error refuses too.
- GOV-03An unanswered request expires after 24 hours and is refused.
- GOV-04Every action record holds 16 fields.
- GOV-05An auditor can verify a record offline, with one command, without trusting elan1.
Level 3 · Attested
Someone independent has examined it.
A third party has audited the claim and put its name to the result.
No elan1 Govern claim is at this level today.
In the platform
The middle layer every action passes through.
elan1 Work and elan1 Decide propose. elan1 Govern decides, so no product builds its own approvals or records. It runs on elan1 Run and works with agents and systems from other vendors too.
“A model may propose an action. It may never approve one.”
- Layer
- Governance
- Platform
- elan1 Govern: Agent governance and assurance
Propose
Applications
Decide
Governanceelan1 GovernAllow · ask a person · refuseCarry
Runtimeelan1 RunIn your country, on your cloudIndustries and use cases
Where elan1 Govern starts.
Banking and financial services
RBI's draft model-risk guidance asks for kill switches, human oversight and traceability.
Insurance
Evidence for every claim and refund action, and for AI-liability cover.
Healthcare and life sciences
Personal-data access and outbound messages, with limits and a named approver.
Also starting in manufacturing: the CFO office's invoices, payments and journals, on a record your auditor can check.
Explore all industriesStatus
Where elan1 Govern stands.
elan1 Govern is pre-release. It has not shipped yet, and we are inviting design partners to shape it on one real process each. The claims register gives the stage of each capability on this page and says when each one becomes true.
40%
A refusal costs 40% of one governed action.
Product fact (pricing)
3
outcomes for every action: allow, ask a person, refuse.
Product fact
24 hours
before an unanswered request expires and is refused.
Product fact
18%
of Indian regulated entities keep AI audit logs today.
Market: RBI FREE-AI, August 2025
FAQs
Questions, answered plainly.
What buyers ask first about elan1 Govern.
Does elan1 Govern replace our approvers?
No. It sends the right actions to the people you name and records what they decide. Routine actions you have already allowed go through without anyone waiting. Every outcome, approved or refused, leaves a signed record your auditor can verify offline, without trusting elan1.
Which AI agents and systems does elan1 Govern work with?
It works with agents built on Claude, the OpenAI Agents SDK, LangChain, Cursor or your own code, and with tools connected through standard agent connectors. Gmail and Zoho Books are built in today. It governs agents from other vendors too, not only elan1 Work and elan1 Decide.
What happens if elan1 Govern is unavailable?
Actions stop; they do not slip through. An error refuses, and the work waits for a person. The design rule is to fail toward a human, never toward an action. Requests left unanswered for 24 hours are refused, and one emergency stop pauses every agent at once.
Why does a refusal cost anything?
A refusal goes through the same checks and leaves the same record as an action that goes through. We bill it at 40% of a governed action, and we publish refusal rates next to cost. Quotes start from one process, in Indian rupees, in writing.
What is human-in-the-loop approval for AI agents?
It means a person must say yes before an agent's risky action runs. In elan1 Govern, actions such as large payments, bank-detail changes and deletions wait for the person you name, who answers in the console or with one tap in Slack or WhatsApp. Unanswered requests are refused after 24 hours.
Does AI agent governance help with the EU AI Act?
The EU AI Act's high-risk duties, which apply to Annex III systems from 2 December 2027, include logging and human oversight. elan1 Govern is designed to supply evidence for both: a signed record of every action and a named person on risky ones. It supports your compliance work; it does not make you compliant.
One agent. On the record.
Start with one process. Add the next when it holds.
Intelligence you can answer for.