Split Tunneling: When to Bypass Your VPN and When Not To
Split tunneling lets you route some apps through your VPN and others directly. It saves bandwidth, fixes streaming-app weirdness, and lets you use your home printer while on a work VPN. It also creates real security trade-offs if you misconfigure it. This is the practical guide: what it is, when to use it, when to leave it off, and how every major VPN implements it.
What split tunneling actually does
By default, a VPN routes everything through the encrypted tunnel. Your work email, your Netflix stream, your DNS lookups, your software updates — all go to the VPN exit server first, then to the destination. This is "full tunneling."
Split tunneling reverses some of those routes. Specific apps, URLs, or IP ranges go direct to the internet instead of through the VPN. The split happens at the OS level — the routing table tells the kernel "send the Slack desktop app via the VPN interface, send Spotify via the regular Wi-Fi interface."
The two flavors
- Split-include (the most common): "Only these specific apps/URLs/IPs go through the VPN. Everything else goes direct." Default-out, opt-in to the tunnel.
- Split-exclude (or inverse split tunneling): "Everything goes through the VPN. Except these specific apps/URLs/IPs go direct." Default-in, opt-out of the tunnel.
For most consumer scenarios, split-exclude is the safer default — everything is protected unless you explicitly carve out exceptions. For enterprise deployments, split-include is more common — only corporate apps go through the corporate VPN, everything else uses the user's direct connection.
The three granularities
- App-based: route by which process initiated the connection. "Firefox goes through VPN, Slack doesn't." The most user-friendly. Works on Windows, Android, modern macOS (via NetworkExtension framework), and Linux (via netns or eBPF).
- URL/domain-based: route by the destination hostname. "work-app.company.com via VPN, everything else direct." Requires the VPN client to integrate with DNS to know which domain you're connecting to before the tunnel decision.
- IP/CIDR-based: route by destination IP. "10.0.0.0/8 via VPN, everything else direct." The original and lowest-level method. Works everywhere but requires knowing IP ranges in advance.
When split tunneling makes sense
Streaming and gaming local content
Many streaming services and games perform worse over a VPN — added latency, occasional IP blocks. Split tunneling lets Netflix or Steam connect direct while your privacy-sensitive apps (browser, messengers) stay tunneled. Common consumer use case.
Home network device access
If you're connected to a VPN, your laptop can't see your home printer, NAS, or smart-home devices on 192.168.1.x. Excluding the local subnet from the tunnel restores LAN access without compromising the VPN's privacy on internet traffic.
Banking and apps that block VPNs
Some banks, payment apps, and government services block VPN IPs as anti-fraud. Split tunneling lets you carve out just those apps to go direct while everything else stays protected. (Better than turning the VPN off entirely.)
Enterprise: M365 split tunnel
Microsoft publishes an updated list of M365 endpoints (Teams, Outlook, OneDrive, SharePoint) and explicitly recommends split-excluding them from corporate VPNs. The volume of M365 traffic from a remote workforce would otherwise saturate the VPN concentrator. Direct-to-M365 routing relieves the bottleneck without compromising security because M365 has its own modern authentication and encryption.
Bandwidth-heavy non-sensitive traffic
Software updates, game downloads, video conferencing to public services. Tunnelling these wastes VPN-provider bandwidth and adds latency for no privacy benefit.
When NOT to split-tunnel
- If you don't understand exactly which apps are routing where. Misconfigured split tunnels expose what you think is being protected.
- If you're in a hostile network and want comprehensive privacy. Public Wi-Fi, hotel networks, censored countries — full tunneling is safer.
- If you're trying to defeat censorship. Anything that goes direct is visible to the censor; the whole point of the VPN is to hide everything.
- If you're using a VPN for an anonymity-sensitive workflow. Even a single direct connection (DNS lookup, browser plugin, telemetry) leaks identity.
Security trade-offs
Split tunneling deliberately punches holes in the VPN's protection. Each exclusion is:
- A connection that bypasses the VPN's DNS protection.
- A connection visible to whatever network you're physically on.
- A connection that reveals your real IP to its destination.
- An attack surface — if a malicious site is somehow in the bypass list, your real IP gets exposed.
The standard advice: prefer split-exclude over split-include (safer default), keep the exclusion list short, audit it periodically.
Implementation across major VPNs
- NordVPN: Split Tunneling on Windows + Android, app-based. Limited on macOS.
- ExpressVPN: per-app on all major platforms. The Windows split tunnelling had a notable DNS-leak bug from May 2022 to February 2024.
- Surfshark: "Bypasser" — app-based and URL-based on Windows and Android.
- PIA: app-based and IP-based on Windows, macOS, Linux, Android.
- Mullvad: command-line/CLI split tunneling on all platforms; less polished UI.
- ProtonVPN: split tunneling on all major platforms, app-based and IP-based.
- Windscribe: built-in browser-extension split tunneling that's particularly granular.
How to test that your split tunnel works as intended
- Connect to your VPN with split tunneling configured.
- Open a browser that's in the tunnel — visit our home page and verify your IP shows the VPN exit.
- Open a browser or app that's excluded — verify it shows your real IP.
- Run our DNS leak test in both — DNS should follow the same routing as the traffic.
- Run our full VPN leak test to confirm WebRTC and IPv6 leaks aren't bypassing your config.
Surprising configuration errors are common. The ExpressVPN 21-month DNS-leak bug only affected split-tunnel users specifically — testing would have caught it earlier than the eventual disclosure.
Frequently asked questions
- Is split tunneling safe?
- It's a trade-off, not inherently safe or unsafe. Every excluded app or URL bypasses the VPN's privacy protections. For non-sensitive traffic (streaming, software updates, home network access) the trade-off is fine and saves bandwidth. For sensitive traffic, full tunneling is safer.
- Should I use split-include or split-exclude?
- For consumers, split-exclude is generally safer — everything is protected by default, you opt out specific apps that need direct access. For enterprises with a defined corporate-app list, split-include is common — only the corporate apps go through the corporate VPN.
- Does split tunneling let me watch Netflix faster?
- Often yes. Streaming services frequently throttle or block VPN connections, and even when they don't, the extra latency of the VPN hop can degrade streaming quality. Excluding the streaming app from the tunnel lets it connect direct at full speed.
- Can I split-tunnel by website?
- Some VPNs (Surfshark's Bypasser, NordVPN on certain platforms, browser extensions) support URL or domain-based split tunneling. Others only support app-level (the whole browser process is in or out of the tunnel). Check the specific VPN's documentation.
- Why did the ExpressVPN DNS leak bug happen?
- The Windows split-tunneling implementation had a flaw that, between May 2022 and February 2024, leaked DNS queries from apps that were supposed to be excluded from the tunnel. The bug specifically affected the interaction between split tunneling and the system DNS resolver. ExpressVPN patched it in early 2024. The episode is the canonical case study for why split-tunnel users should periodically verify their setup with a leak test.