Site-to-site and remote-access VPNs: two tunnels, two trust models
A tunnel protects the path, not the endpoint: the device at the far end is unverified.

A virtual private network (VPN) is a virtual network built on top of existing networks, carrying traffic between two points inside a protected envelope: confidentiality, integrity, data origin authentication, replay protection and access control, applied at the network layer (nist-sp800-77r1.txt lines 1441–1442; rfc4301.txt lines 262–267). A VPN does not change the trust status of the endpoints. The tunnel decides which traffic is protected and who may open it — not whether the host at the far end deserves the access it receives. The two architectures that carry almost every deployment — site-to-site between two gateways, remote access between a gateway and a device nobody manages — differ less in cryptography than in where that decision sits.
What a tunnel protects, and where it stops
IPsec adds traffic analysis protection and access restrictions on top of confidentiality, integrity and mutual authentication (rfc7296.txt lines 287–288; nist-sp800-77r1.txt lines 1434–1437). The limit is geometric: the protection spans the tunnel and stops at its endpoints.
site-to-site
host ──[ unprotected ]── gateway A ═══[ protected ]═══ gateway B ──[ unprotected ]── host
remote access
laptop ═══[ protected ]═══ VPN gateway ──[ unprotected ]── internal servers
SP 800-77 Rev. 1 draws the same picture for both: the gateway-to-gateway architecture “only protects data between the two gateways”, the remote access VPN “does not provide full protection for data throughout transit” (nist-sp800-77r1.txt lines 1163–1164, 1229–1230). A VPN shortens the exposed path; it does not remove it.
The pieces an IPsec tunnel is made of
- SA — Security Association. “A simplex "connection" that affords security services to the traffic carried by it” and “a management construct used to enforce security policy for traffic crossing the IPsec boundary” (
rfc4301.txtlines 629–630, 1050–1051); one-way, so a conversation needs a pair; its scope is the negotiated traffic selectors (nist-sp800-77r1.txtlines 1667–1674). - ESP — Encapsulating Security Payload. “Confidentiality, data origin authentication, connectionless integrity, an anti-replay service … and limited traffic flow confidentiality” (
rfc4303.txtlines 31–34). AH is integrity only and is discouraged (nist-sp800-77r1.txtlines 2695, 2752). - Mode. Tunnel mode wraps the original packet behind a new IP header, so it serves both architectures and survives NAT (
nist-sp800-77r1.txtlines 2311–2317, 2339–2341). - IKE — Internet Key Exchange. “IKE performs mutual authentication between two parties and establishes an IKE Security Association (SA)” (
rfc7296.txtlines 287–288). The authentication method is the trust anchor: a group pre-shared key, once obtained, lets an attacker impersonate “any client or the remote access VPN server itself” (nist-sp800-77r1.txtlines 1896–1897).
Site-to-site: two gateways and the networks behind them
Two gateways hold one IPsec connection, each side’s networks sit behind its own gateway, and hosts reach the other side without knowing a tunnel exists — SP 800-77 Rev. 1 calls this “the easiest to implement in terms of user and host management” (nist-sp800-77r1.txt lines 1168–1170).
Both endpoints are managed by the organisation, and the addresses behind them are known. The tunnel is a transport for a routing relationship, so the reachable prefixes have to be planned and filtered deliberately, and what arrives through it looks trusted unless something checks it again. IPsec access control “works independently from other access control mechanisms, such as firewall services” (nist-sp800-77r1.txt lines 946–947); a firewall ruleset still has to exist alongside it (nist-sp800-77r1.txt lines 4186–4187).
Remote access: a device you do not manage
The remote access architecture inverts the assumption. A gateway faces the internet, each remote device opens its own connection to it, and the user authenticates before the tunnel is usable (nist-sp800-77r1.txt lines 1186–1195). The endpoint is now a laptop or a home computer nobody manages: the tunnel extends the internal network to a machine whose state is unverified.
Remote access comes as a network client or as a browser session over TLS (nist-sp800-113.txt lines 661–666, 684–694). Both leave the endpoint unverified: for endpoints outside the organisation’s control, “malicious activity could then enter the organization’s networks from the endpoints through their IPsec connections” (nist-sp800-77r1.txt lines 4194–4199).
Endpoint security controls are a separate control: host integrity is checked before and during the session, granting full access to a managed laptop and limited access to a personal one (nist-sp800-113.txt lines 1156–1166); SP 800-77 Rev. 1 offers the harder option, to “automatically quarantine each remote host that establishes an IPsec connection” (nist-sp800-77r1.txt lines 4211–4214). That is a feature beside the tunnel, not a property of IPsec.
Split tunnelling is the other half of the question: when the client does not send all its traffic through the gateway, the host is on the organisation’s network and on another one at once, and a compromised host can be used to reach the organisation through its own tunnel (nist-sp800-77r1.txt lines 3664–3666). It is “strongly discouraged” (line 3697): a policy trade-off, not a checkbox.
The common mistake: a tunnel is not a zone
“The VPN is our secure network” collapses two different things. A VPN is a protected path plus an authentication gate; a trust domain is a set of systems that share a security requirement. Putting a remote device inside the internal address space does not enrol it in that domain, and treating the tunnel as the boundary produces exactly that: malicious traffic entering “through their IPsec connections” on a host nobody inspected. A VPN that is the only control at a boundary is an encrypted door with no lock behind it.
Level and prerequisites
L3. It assumes the L1 vocabulary of packet forwarding and the IP header, and the L2 material on routing and the next hop.
Where to go next
References
- RFC 4301 — Security Architecture for the Internet Protocol (Kent, Seo, December 2005): the security services, the SA definition and its one-way nature, the SA as a policy construct.
- RFC 4303 — IP Encapsulating Security Payload (ESP) (Kent, December 2005): the services ESP provides.
- RFC 7296 — Internet Key Exchange Protocol Version 2 (IKEv2) (Kaufman, Hoffman, Nir, Eronen, Kivinen, October 2014): mutual authentication, the IKE SA and the child SAs.
- RFC 8446 — The Transport Layer Security (TLS) Protocol Version 1.3 (Rescorla, August 2018): TLS as the transport on which the TLS-based VPN family runs.
- NIST SP 800-77 Revision 1 — Guide to IPsec VPNs (June 2020): the VPN definition; the gateway-to-gateway and remote access architectures and their partial protection; the IPsec components and modes; traffic selectors; split tunnelling; the endpoint outside the organisation’s control and the quarantine option.
- NIST SP 800-113 — Guide to SSL VPNs (July 2008): SSL portal and SSL tunnel VPNs, the TLS port used by the portal, and endpoint security controls.