Controls & Monitoring
AI Agent Risk Monitor
Monitors agent actions for risky or unexpected behavior.
Controls & Monitoring
AI Agent Permission System
What each AI agent is allowed to do, action by action, is defined and enforced, with approval required for anything sensitive.
An AI agent that can send emails, update records or call outside services needs clear limits. This system lists every action each agent can take and sets a rule for each one: run automatically, wait for a named person to approve, or never run. Rules are enforced the moment an agent tries to act, and every attempt is logged.
- Every agent action set to allow, approve or block
- Sensitive actions routed to a named approver in Slack or Jira
- A full log of what each agent attempted and why
$5,400
20–30 working days
Optional maintenance $270/mo
What it is
AI Agent Permission System is an agent that catalogs every action your AI agents can take, such as sending a message or writing to a record, and enforces explicit rules for which actions run automatically, need approval, or are blocked. It governs agent behavior specifically, not human or system access.
Why businesses buy it
Agents are useful because they act on their own, and that is also what makes them risky. One wrong email to a customer list or one deleted record can cost more than the agent ever saved. Most teams rely on the agent's own judgment, which is not a control. This system puts your rules between the agent and the action, so permissions grow only when someone decides they should. The written scope sets the agents, actions and systems covered before you pay.
Problems this solves
AI agents can send messages or change records without any check
Every action your agents can take is cataloged and given a rule: run automatically, need approval, or blocked. Up to five controls are configured across up to three agents or applications.
Nobody has a full list of what each agent can do
The action catalog becomes a written record in the admin console or report. You can see which actions exist for each agent and which rule applies to each one.
Approvals for risky agent actions happen informally, if at all
Actions marked for approval pause until a named person signs off. The workflow and guardrail configuration sets who approves and what the agent does while it waits.
A good fit if
Your AI agents can take actions, such as sending emails, updating records or calling external APIs, and you want limits on what they may do unsupervised.
You want a clear, written line between actions an agent can take automatically and actions that need a person's approval first.
You are expanding what your agents are allowed to do and want that expansion to happen through explicit rule changes, not silent drift.
You want every action attempt logged with its outcome, so you can see what your agents actually did, not just what they were built to do.
Not the right fit if
Your concern is who, people or systems, can access your AI applications broadly, not agent actions specifically; see AI Access Control System.
You want this combined with input screening and centralized logging across multiple agents in one build; see AI Agent Security Layer.
Your priority is malicious input reaching your agents, not the actions they are permitted to take; see AI Prompt Injection Defense.
You need written policy documents defining acceptable agent behavior first, before enforcing them; see AI Security Policy System.
How it works
- We connect to the AI agents and tools you name. Every action an agent can call, such as sending a message, updating a record, or hitting an external API, is cataloged as part of setup.
- Permission rules are set per action, per agent. Together we decide which actions run automatically, which need a logged approval from a named person, and which are blocked outright regardless of context.
- Rules are enforced at the point an agent attempts an action. Rather than trusting the agent's own judgment, the permission system checks the rule before the action executes, every time.
- Approval requests route to a named person. When an agent attempts something that requires sign-off, the request reaches Slack or Jira with enough context for a quick, informed decision.
- Every attempt and its outcome are logged. Whether an action ran automatically, waited for approval, or was blocked, the record shows what was attempted and what happened, for every agent in scope.
- You get a test report and a working rule set at launch. The 60-minute handoff covers how to add a new action or adjust an existing rule as your agents take on more work.
What you provide within 72 hours
Your portal lists these the moment you order. Delivery starts the working day after everything is in and your access is checked.
- Written authorization confirming which AI agents and systems we may govern.
- The list of AI agents and the actions each one is able to take.
- Your approval rules for actions that should require a named person to sign off.
- Read and write access to the agents' tool-calling and messaging pipeline, as needed.
- A named IT or security owner who can handle approval requests after launch.
- Admin access and API keys for every connected system, shared by adding our team as a user or creating a key you control, never by email.
- One decision-maker who can approve the permission rules and sign off the launch.
The 72-hour rule. Send everything on your checklist, including access, within 72 hours of ordering so your build starts on time.
Timeline
Start
Agents, actions and permission rules confirmed.
Weeks 1–2
Permission enforcement and approval routing built.
Week 3
Agent, tool and alerting connections completed.
Weeks 3–4
Documented test scenarios run and results sent to you.
Week 4
Your review and two rounds of changes to the rules.
Go-live
Live, recorded handoff session, 30 days of support begin.
| Option | Delivery time | Price |
|---|---|---|
| Standard | 20–30 working days | $5,400 |
| Express | 15–25 working days | $6,200 |
| Rush | 10–20 working days | $7,000 |
What you receive at handover
- AI Agent Permission System configured and deployed, covering up to three agents or AI applications
- A catalog of agent actions, each set to run, need approval or be blocked (up to five controls)
- Workflow, prompt and guardrail configuration, with connections to the agreed systems
- A test report with a control test log for the agreed scenarios
- An admin guide and a recorded 60-minute handoff session
- 30 days of post-launch support
Expected benefits and how to measure them
- Can let you hand agents more work while sensitive actions stay under human control.
- Helps you explain to leadership exactly what each agent is allowed to do.
- Makes it easier to tighten or relax a rule without rebuilding the agent.
- Approval requests and rejections
- Count approval requests per week in the console, and how many were rejected. Rising rejections can show an agent needs better instructions.
- Blocked actions per agent
- Count blocked attempts per agent each month. Review whether the rule or the agent's prompt needs changing.
- Sensitive actions run without sign-off
- Compare your agent logs from the month before launch with the console record for the month after.
Before you start: List each agent in scope and every action it can take today. Mark which of those actions currently run without anyone checking.
These are things to track, not promised results. Outcomes depend on your data, your team and your market.
Running costs
The build price is one-time. Running the permission system costs roughly $20 to $300 a month in your own accounts, for hosting, logging and the monitoring tools it relies on day to day.
AI model usage during the build and testing is billed the same way, directly to your accounts, so you keep ownership of every provider relationship involved.
Ask about running costsCustom requirements
If you need something outside the Scope Card, such as permission rules across more than three systems, add a note at checkout. You get a written Change Quote, and that work only starts once you accept it. Larger or bespoke projects go through the Custom AI request form instead.
Request a custom versionKeep it running
We recommend monthly maintenance, which covers monitoring and a monthly allowance of change hours for things such as adding a new action or adjusting approval rules. It starts after the 30 days of included support and can be cancelled any month.
$270/mo (5% of the build price a month, includes 2 change hours)
Optional. Starts after the post-launch support period. Cancel any month.
How maintenance worksHow this compares
| Do it yourself with a SaaS tool | Hire a custom agency | Upmark Labs | |
|---|---|---|---|
| Price | An agent orchestration platform bundles permissions into a broader monthly fee | A security firm quotes after a scoping call | Ours is one fixed price, shown on this page before you buy |
| Scope | An agent orchestration platform's permission model is fixed by its design | An agency agrees scope during the engagement | Ours is written per action, per agent, on the Scope Card |
| Delivery | An agent orchestration platform still needs configuration to be useful | A security firm engagement often takes weeks to schedule | Ours has a fixed delivery time shown before you buy, with faster options |
| Approval routing | An agent orchestration platform may lack a built-in approval step | An agency build depends on what was agreed | Ours routes approval requests to Slack or Jira automatically |
| Certification | An agent orchestration platform does not certify anything on its own | An agency may offer certification as a separate service | Ours enforces the permissions you agree; no certification is implied |
Questions about this product
What does AI Agent Permission System actually control?
It controls which specific actions your AI agents are allowed to take, such as sending a message, updating a record, or calling an external API, cataloged action by action. Each one is set to run automatically, require a named person's approval, or be blocked outright, enforced every time it is attempted.
How is AI Agent Permission System different from AI Access Control System?
AI Agent Permission System governs what actions your AI agents themselves may take once they are already running. AI Access Control System governs who, people or connected systems, can access your AI applications and data in the first place. One is about agent behavior; the other is about entry.
How much does AI Agent Permission System cost to run each month?
Expect roughly $20 to $300 a month in your own accounts for hosting, logging and the monitoring tools the permission rules depend on. The figure moves with how many agents you run and how often they act. You pay those providers directly, with no per-action fee from us. The one-time build price for each delivery speed appears in the price panel on this page.
What happens when an agent attempts a blocked action?
The action does not execute, and the attempt is logged with what was tried and which rule stopped it. If the action instead requires approval rather than an outright block, the request routes to a named person in Slack or Jira for a quick, well-informed decision. This keeps a clear, auditable trail of every action your agents have attempted, approved or otherwise.
Can an agent's permissions expand as it takes on more work?
Yes, but only through an explicit rule change your named approver makes, not automatically as the agent encounters new situations on its own. That keeps permission growth deliberate and fully logged, rather than an agent quietly gaining capability nobody agreed to. This keeps trust in what your agents can do deliberate, not something that erodes quietly.
Can I add a new agent or action later?
Yes, within the three systems included in scope, adding an agent or a new action fits inside the monthly change hours that come with maintenance. Extending coverage beyond three systems comes through a Change Quote, priced and agreed with you before that work starts. This keeps the cost of expanding coverage predictable as your agent footprint grows over time.
Who owns the permission rules and action log once it is live?
You do. The permission rules, the approval routing and the full action log all live in your own connected accounts and tools, not a separate platform we control, so you retain complete visibility into everything your agents have done. We keep no separate copy that would lock you into working with us going forward.
Decide what your agents can do before they do it.
Order AI Agent Permission System and action-by-action rules start enforcing across your agents, live on a fixed delivery date.
Works well with
Controls & Monitoring
AI Security Policy System
Maintains security policies and checks systems against them.
Controls & Monitoring
AI Security Monitoring
Centralizes security events from AI systems with alerting.