Your agents leave as anonymous cloud IPs.Fix that.

Your agents reason confidently about infrastructure they cannot see, and leave as anonymous cloud IPs you cannot tell apart.

1An address per agent
agent-01software agentlaptop-1182hostbuild-runner-7containerone /128 each, on AS219419
2Policy at its egress
IFidentity = agent-01dest registered < 30 daysDENY
3Cut from off the host
agent-01revokedegressclosedhost agentnot askedone command, from the control plane

Five questions your stack cannot answer.

Your agents share a cloud range and carry tokens that leak. Nothing in the stack can say what any one of them may reach.

How many agents are running in here?

Nobody filed a ticket for the ones that matter.

Who started this one?

The identity is in the payload. The connection carries none.

What is it allowed to reach?

Your allowlist is addresses, and the infrastructure moves.

Can you prove it never touched anything sanctioned?

You have logs of what you saw. You have no proof of what was prevented.

Can you switch off one agent and leave the rest?

Not while they share an address.

An identity, and a leash, for every agent.

  1. One address per agent

    A routable IPv6 address from our own RIPE-allocated space, RPKI signed, with forward and reverse DNS in step.

    See Whisper Graph XDR

  2. Answers it can ground

    Any MCP client queries real infrastructure mid-task, and can tell no data from known clean.

    Set up MCP

  3. One call to revoke

    Kill the identity and close its egress. The agent is off the network whether or not it still runs.

Anyone can check it.

A third party checks one of your agents with dig, whois and openssl. No account, and no Whisper service in the trust path.

  • Tied to whoever launched it

    Every identity traces back to the person or system that created it, with the tree for sub-agents.

  • Policy in the terms you think in

    Rules on sanctions, geography, Tor exits, new registrations and RPKI, applied at resolution and at egress.

  • Deployed with what you run

    Intune, Jamf, Kandji, Omnissa or Hexnode, or a universal installer that does not care which.

forward-confirmed DNSpassDANE TLSApassDNSSEC chainpassRDAP registrationpassshared secret requirednoneanyone can re-run these without our cooperation

Where to start building.

This job is set up once, so each card opens the guide for its part.

Check it yourself.

  • Our own network

    AS219419: RIPE allocated, RPKI signed and MANRS compliant.

    stat.ripe.net/AS219419

  • Every kind of identity

    Host, mobile, software agent, sub-agent, BYOD, OT and IoT.

  • Three tools to verify

    dig, whois and openssl. You already have all three.

  • For a crawler operator

    Every request arrives from an address whose registry record names the operator and the abuse contact. An origin checks that at its own edge, without calling us.

Write down the rule you could never enforce.

Give your first agent an identity, or walk us through your fleet on a call.