What zero trust means
Older security worked like an office building: once you're past the front desk, you can walk anywhere. Zero trust works like a building where every door checks your badge, every time, and your badge only opens the rooms you need today.
Nobody is trusted because of where they are or what they did a minute ago. Every request has to prove it's allowed.
Why AI agents need it
An AI agent is a new kind of colleague. It's fast, it doesn't get tired, and it does what it reads. That last part is the problem.
It reads strangers' text
Emails, PDFs and web pages can hide instructions. An agent can't always tell your instructions from someone else's.
It can act
An agent with access to your accounting, inbox and CRM can send, pay and change things, not just answer questions.
It works at scale
One mistake can be repeated across hundreds of invoices or tickets before a person looks.
You can't fix this by telling the agent to be careful. You fix it by building a system where a tricked agent still can't do harm. That's what zero trust gives an AI workforce.
Three principles
- 1
Check every request
Each request an agent makes is checked against the task's rules: which site, which tool, which team's data. Nothing passes because an earlier request did.
- 2
Give the least access
A task gets only the tools it needs, a stand-in instead of your real keys, and a budget. When the task ends, the access ends.
- 3
Assume something will go wrong
One day an agent will read a malicious document. The design assumes that, so the damage stops at the workspace wall and a person decides what happens next.
How it's built in
These aren't settings you switch on. Every agent on Upmo AI works this way from its first task.
Agents only reach sites you allow
Each task runs in a sealed workspace with no internet access. It can only reach the tools on its own list.
Read more → KeysAgents never hold your keys
Agents work with stand-in keys. Upmo AI adds the real one on the way out, only for allowed sites.
Read more → ApprovalsRisky actions wait for a person
Sending data out, moving money or changing records stops and asks. You see exactly what will happen.
Read more → DataEach team only sees its own data
Access rules are enforced by the database itself, so a bug in the app can't show one team another's records.
Read more → Sign-inSign in with a passkey or SSO, and confirm again for big changes
Use your fingerprint or face, or your company's SAML single sign-on. Changing roles, keys or security rules asks you to confirm again.
Read more → RecordsEvery action and every memory has a source
Each step is logged, and each fact an agent remembers shows where it came from. Correct or delete it at any time.
Read more →What we don't claim
Upmo AI isn't SOC 2 or ISO 27001 certified yet. No system is perfectly secure, and we won't tell you ours is. What we can offer is this: the code is open, so your security team can read exactly how each of these controls works.
Found a problem? See our Security Policy or email security@upmo.ai.