14:01 · Delegation active
RECEIPT · r-8812
$12,400 → Northwind Supply
- Authority
- current
- Revocation
- not revoked
Executed
AI agent controls
Did your AI do what it was told?
Inherence stops any action that breaks your rules. Every action carries proof your team can check.
Our team comes from
- MIT
- Jane Street
- Columbia
- MITRE
- Harvard
- Snap
The problem
You can see opportunities AI could enable in financial and health care sectors. Without a reliable control mechanism, you can’t tap any of them.
Where money or sensitive data is involved, consequential actions need to follow your rules.
- INFOagent=treasury-07 task="rebalance client hedge book"Strategy exposed
- DEBUGsystem_prompt="Short the ETH perp when funding exceeds 8 bps. Target leverage 3.2x."
- TRACEdb.read row={name:"Dana Whitfield", acct:"****4417", balance_usd:482913.22}Customer data revealed
- DEBUGauth.grant token=ak_live_7f3ea91cc402 scope=*
- WARNpolicy.check=skipped
- INFOtransfer.out amount_usd=1010736.00 approver=none
- INFOlog.flush delay=47sMoney already moved
How Inherence helps
Stop problems before they happen. Apply AI agent authorization at the action boundary.
Every action carries proof that anyone can verify. So you can turn agents loose with confidence that they will obey your rules without exposing sensitive data.
Agent proposes trade
Inherence
Executes:
Within mandate
Within mandate
Blocked:
Off-mandate
Off-mandate
Illustrative workflow
An agent requests a payment.
Inherence checks whether the request satisfies your rules.
treasury-agent-07 requests$12,400 → Northwind Supply
- Payee allowlist
- Per-action cap
- Daily budget
- Delegation active
A compliant request proceeds through the authorization path. A failed request is blocked or escalated. The receipt identifies the policy version and request used for the decision.
INHERENCE- Executes:
Within policy - Blocked:
Over daily budget - Escalated:
Awaiting owner
RECEIPT
- Request · r-8812
- Policy · v3.1
- Hash · a7f3…c21e
The account owner or downstream service receives the receipt. Current authority and revocation checks remain part of the authorization design.
14:02 · Delegation revoked
RECEIPT · r-8813
$3,150 → Northwind Supply
- Authority
- revoked
- Revocation
- 14:02 by owner
Blocked
For investment-specific delegation, explore Investment Mandate Controls. For multi-party payment flows, explore Payment and Treasury Controls.
Inherence proof
- Input
- : 1.20550000 BTC
- Output #1
- : bc1qx9a...7k4dl
- Output #2
- : bc1q4f...n2ze7
- Network fee
- : 61.8K sats
- Confirmations
- : 6
- Enforcement verified
- Privacy maintained
- Processed instantly
Why Inherence
Control
Stop rule-breaking actions before they happen. Write your rules in plain English. Inherence turns them into a check that runs before every action.
Proof
Every action comes with a proof (receipt) that anyone you work with can check themselves.
Privacy
Prove the rules held without sharing your data. Amounts, limits, and strategies stay private. The receipt shows the rules held and nothing more.
Expand your capabilities and breathe easy.
Get your AI agent working for you.
Agent security is moving toward deterministic authorization.
NIST's October 2026 agent identity design tenets describe separation of policy from reasoning, deterministic decisions, and evidence maintained independently of the agent. These principles reinforce the need for explicit controls.
FAQs
AI agent controls are the rules an agent must follow before it can act: approved vendors and tools, spending caps, budgets shared across a fleet, and required sign-offs.
Guardrails and wallet limits work inside one tool or one wallet, and everyone else has to trust them. Inherence applies your rules across the tools and payment rails you route through it, and every action carries a proof anyone you choose can verify.
The agent may still propose the action, but it can't execute if it breaks your rules. If a hidden instruction tells an agent to pay a new wallet, the payment is blocked because the wallet isn't on the approved list.
No. Inherence sits between the agent and the wallet, payment rail, or tool it acts through. Your agents, models, and providers stay the same.
It checks whether an action stayed within the rules you set. Whether the purchase was a good idea is still your call.
Only actions routed through Inherence are checked. To cover an agent's spending fully, its wallet or payment path needs to require the check before it signs.
Yes. Rules can apply to a single agent or to every agent you run, including shared budgets and limits that add up over time.
Yes. You can start by running past agent actions through Inherence, or by checking live actions without blocking them. Blocking requires Inherence to sit in the path that signs or sends payments.
Request an Agent Controls Review
What do you need your AI agent to do? Describe the action and the boundary that must hold. We will follow up to assess the policy, integration point, and evidence needed for a focused evaluation.