- Agent
- Calendar assistant
- On behalf of
- Laura · Finance
- Purpose
- Plan tomorrow's meetings
- Wants to
- Read calendar: titles, times
Your AI. Your Data. Your Rules.
Control every AI action. See what happened. Stop what should not happen.
SafeUs checks who an AI agent is, what it wants to do, which data and tool it wants to use, and whether approval is required — before the action happens. Then it records what actually happened with evidence.
SafeUs sees, checks and controls supported AI actions — and proves what happened.
AI Passport ControlInteractive demo
- Identity
- Delegation
- Purpose & data
- Tool & policy
Written to Black Box
See what an AI agent can reach, and what is waiting for you
Follow one example agent through SafeUs: review its access, answer a request that needs a person, withdraw a permission and check the result.
Access
- Read shared reportsAllowed
- Draft summariesAllowed
- Send files outside the teamNeeds approval
- Delete filesNot granted
Request
Send 3 reports to an outside address
- Allow
- Ask a person
- Refuse
Record
- Send outside: held for approval
- Declined by you
- Send outside: access withdrawn
Access now: Read shared reports · Draft summaries
Recheck: the next send outside the team is refused before it runs.
Step 1 of 5 · Register the agent. The agent and who it acts for appear in SafeUs.
The agent and who it acts for appear in SafeUs.
Each permission shows its state: allowed, needs approval or not granted.
Sending files outside the team needs approval, so the request stops and waits for you.
You decline. The answer and the rule that applied are recorded.
You withdraw the send permission. A recheck confirms what the agent can still do.
What SafeUs protects
Everything AI can reach, in one view
Phones, laptops, mail, files, calendars, chats, social accounts, websites — and the AI agents that want to use them. SafeUs sits in the middle and decides what each one may do.
Devices · Available today
Phone
See which apps and AI assistants on your phone can reach your accounts. Planned: one tap to cut a lost phone's access.
Mail assistant asks to reach Phone → SafeUs checks first
Tap any icon to see what SafeUs does there.
One control layer. Four capabilities.
What SafeUs does, at a glance.
AI Action Control
Before an AI acts, SafeUs checks the rules.
- Who
- Purpose
- Data
- Tool
- Policy
- Action
AllowApprovalHoldDeny
Show details
Every request is checked in a fixed order. A missing answer never becomes a yes.
AI Black Box
Every controlled AI action leaves a verifiable trail.
- Who authorized it
- What was requested
- Which rule applied
Show details
- Human approval, if required
- What actually happened
- Evidence status
Shield
Contain a suspicious agent in one step.
- Under Watch
- Quarantined
- Blocked
Safety Check-up & Emergency Stop
Check now. Stop fast when something looks wrong.
Show details
Safety Check-up checks without destructive changes. Emergency Stop previews the impact first, then pauses/revokes only what SafeUs can actually control.
Simulation
See how SafeUs checks an AI agent's request
Pick an example. Follow a request from an AI agent through the passport check and see whether SafeUs lets it through, holds it for a person or stops it before it reaches your accounts.
Simulation · illustrative data, not live traffic
Choose an example
Request path
Hidden instruction
Unregistered agent
Changed tool
Inbox assistant
SafeUs checkpoint
Mail
Files & drive
Calendar
Payments
Chats
Website
Hidden instruction in an email
An email hides a line telling the assistant to send every contract to an outside address.
Inbox assistant · send 40 contract files to an outside address
Agent passport check
- Who the agent is…
- Acting for whom…
- Purpose…
- Allowed access…
- Rules and risk…
SafeUs decision
…
What SafeUs does
- Stops a send that passes through SafeUs before the file leaves
- Quarantines the assistant until you review it
- Shows you the email that carried the instruction
Across these examples
- Requests checked
- 8
- Held for a person
- 3
- Stopped
- 4
- Inbox assistantFiles & drive…
- Scheduling assistantCalendarAllow
- Research agentWebsiteDeny
- Clean-up agentFiles & driveHold
- Support agentChatsHold
Three screens, one proof chain
Protect. Decide. Prove.
See your digital world at a glance, answer the requests that matter, and keep a record you can show anyone. All three use the same rules and the same Black Box.
My SafeUs
Mail2 AI apps can read · sending asks you
Try it: tap the screen
Your digital world
Every account and device AI can reach, with a status you can read in a second.
- Mail, files, calendar, chats and websites in one place
- Green, amber, red: allowed, asks you, stopped
- Lost phone or odd agent: one tap cuts access
Waiting for you · 21:42
Sales copilot wants to send 40 e-mails
Purpose: follow-up after the trade fair
- Identity checked
- Purpose matches the request
- Scope: names and e-mails only
- New recipients → asks you
Checking…
Try it: tap the screen
You decide what matters
When an AI wants to send, delete, pay or share, it waits for you — with one clear sentence, not a log.
- Identity, purpose and scope checked first
- Allow once or deny, from your phone
- Every answer is recorded
Black Box · today
Try it: tap the screen
The Black Box keeps the proof
Who asked, under which rule, who approved and what happened — as references and hashes, never content.
- Tap a line to see the rule and the hash
- Export it for your auditor or insurer
- History kept according to your plan
The control layer for the AI-connected world
AI already works inside your company. Someone should check its papers.
Assistants read mail, agents open files, copilots call your systems — usually with a person's full access. SafeUs turns each of those accesses into a decision you can see, limit and prove.
- SeeWhich AI can reach what?
- One inventory of accounts, agents and connected systems, with how each one is covered — counts and status, never a score.
- ControlCan we stop it right now?
- Allow, limit, or ask a person first. Emergency Stop previews the impact, then pauses what SafeUs controls.
- ProtectWhat did it actually take?
- Every decision is written before anything is released: which fields, for which purpose, approved by whom.
Shield
Watch. Quarantine. Block.
Three clear states for an agent or connection you are not sure about. A person with the right role makes each change, it is recorded, and lifting it requires revalidation.
- Under Watch
Tighter checks: every action other than reading goes to a person.
Typical triggers: tool changed after approval, a suspicious instruction, another service's key, repeated denials.
- Quarantined
Stopped until reviewed. Nothing is released meanwhile.
Lifted only after identity, permission and connection health are revalidated.
- Blocked
Access revoked by an authorized human.
Automation can never lift a block — only a manager can, after revalidation.
Shield only tightens. It never grants more access than your rules already allow.
On a bad day
When something looks wrong, you need more than an alert.
Preview the impact, stop supported access, recheck the environment and get an evidence-backed report — including what SafeUs could not do for you.
- PreviewSee what will change before anything changes.
- SnapshotRecord the current state as references.
- ContainPause supported agents and connections.
- RevokeWithdraw the permissions SafeUs granted.
- RecheckRecheck connections and tools.
- RevalidateConfirm identity, permissions and connection health.
- ProveSeal the report with evidence.
Restore appears only for reversible states. A revoked connection shows “Reconnect required” — never “restored automatically”.
Black Box
The flight recorder for AI actions.
Reconstruct who requested an action, under whose authority, against which rule, with which approval, and what happened afterward.
- AgentSales agent
- RuleCustomer export needs a person
- ApprovalData owner · once
- ActionExport 1 report
- OutcomeCompleted
- EvidenceEvidence Ready
Action recording is available only for connections that pass through SafeUs control points.
One engine. Four scopes.
The same control engine for every size of organisation, not four different products.
Know which supported AI connections can access your digital life. Review and stop suspicious access.
Protect my digital worldSee. Control. Protect.
Start with what you can see today.
Connect one account or one agent. SafeUs shows what can reach it, checks every supported action and keeps the record.
Early access for teams and partners
Enterprise isolation and the partner programme open in waves. Leave your e-mail and we'll contact you.
We use this e-mail only to contact you about access, and delete it after 12 months. Privacy