Every S/4HANA program with field engineers, plant managers, external auditors or a sales force that won't install a VPN client ends up with the same request: the Fiori launchpad, from outside the corporate network. The design usually proposed is to publish the front-end server, put an SAP Web Dispatcher in the DMZ in front of it, buy a certificate and go live.
I think that design is wrong, and not only for the general defense-in-depth reason. What you publish is an ABAP application server, and over the last four years of critical SAP vulnerabilities, the customers who came through unharmed were the ones whose ICM could not be reached from the internet. Patch speed had little to do with it.
What gets exposed
Publishing a Fiori front-end server puts the Internet Communication Manager on the internet, and behind the ICM sits the whole Internet Communication Framework service tree.
Fiori itself needs about a dozen paths: the launchpad shell under /sap/bc/ui2/, UI5 resources under /sap/bc/ui5_ui5/, OData services under /sap/opu/odata/. A default S/4HANA front-end server has hundreds of ICF nodes active. The rest of the tree typically includes:
| Path | What it is |
|---|---|
/sap/bc/gui/sap/its/webgui | SAP GUI in a browser, with the full transaction set behind it |
/sap/bc/soap/rfc | RFC over SOAP |
/sap/bc/webdynpro/ | A separate UI stack with its own CVE history |
/sap/bc/adt | The ABAP development endpoint |
/sap/public/info | SID, kernel release, host name and time zone, with no authentication |
/sap/public/info does the attacker's reconnaissance for them. Matching the kernel patch level against SAP's published notes is then bookkeeping.
The launchpad adds more. OData $metadata documents describe your entity sets, which amounts to a readable map of the business functions the system offers. And unless you have changed it, the login form accepts basic authentication against the SAP user store, with no second factor, at whatever rate the ICM will serve.
The login screen is not the boundary
Two SAP vulnerabilities show why "an attacker still has to log on" is the wrong model.
CVE-2022-22536, which Onapsis named ICMAD, is a memory pipe desynchronization flaw in the ICM. CVSS 10.0, exploitable without authentication with a single crafted HTTP request, leading to session hijacking and system takeover. SAP patched it in February 2022. It affected NetWeaver ABAP and Java, the ABAP Platform, SAP Content Server and the SAP Web Dispatcher. CISA added it to the Known Exploited Vulnerabilities catalog that August.
CVE-2025-31324 is a missing authorization check in the NetWeaver Visual Composer metadata uploader, also CVSS 10.0. An unauthenticated POST to one endpoint uploaded a JSP web shell. SAP issued an out-of-band patch on 24 April 2025. Censys counted more than 7,500 exposed NetWeaver application servers and estimated that most internet-facing Java stacks had the vulnerable component enabled. EclecticIQ later recovered an attacker's own log files listing 581 compromised systems and 1,800 further domains queued for targeting. The activity was attributed to China-nexus groups and hit gas distribution, water utilities, medical device manufacturing and government ministries.
The timeline:
| When | What happened |
|---|---|
| Late January 2025 | Palo Alto Unit 42 observes probing against the vulnerable endpoint |
| From 27 March 2025 | Rapid7 confirms exploitation in the wild |
| 24 April 2025 | SAP publishes an out-of-band security note |
| Weeks after | Censys counts 7,500+ exposed servers; EclecticIQ recovers attacker logs naming 581 compromised systems |
For a month, arguably three, any internet-facing system with that component enabled could be compromised, and there was no patch to apply, no signature to deploy and no advisory to read.
Visual Composer is a Java stack component, so this CVE does not concern the Fiori front-end server as such. ICMAD does: it hit the ICM, which is what you publish when you publish a Fiori server. What carries over from CVE-2025-31324 is the pattern: a pre-authentication remote code execution in an internet-facing SAP web stack, disclosed after weeks of exploitation.
Neither patching discipline nor password policy would have helped in that window. If an unauthenticated attacker cannot open a TCP connection to your ICM, a pre-authentication RCE in the ICM becomes a maintenance ticket instead of an incident.
What the Web Dispatcher does and does not do
The SAP Web Dispatcher belongs in this architecture, but it is not a security boundary on its own. Treating it as one is the design error I see most often here.
| It does | It does not |
|---|---|
| Terminate TLS | Inspect payloads |
| Distribute load across application servers | Rate-limit credential stuffing |
| Hide internal host names and ports | Enforce MFA |
| Filter URLs against a permission table | Evaluate device posture |
| Rewrite headers | Decide who the user is |
| Give the browser a single origin, so the same-origin policy does not break your OData calls | Anything, by default, other than forward |
It also runs the same ICM code as the application servers behind it, which is why ICMAD affected it too. Onapsis noted that CVE-2022-22536 was exploitable specifically where a proxy sits between clients and backend, which is the case in every production landscape. The proxy layer is part of the attack surface.
The architecture
The fourth layer is the one people have trouble naming, which partly explains why it gets left out. The usual term is authentication proxy, or pre-authentication reverse proxy. Vendors sell it as an identity-aware proxy, access gateway, web access management or ZTNA broker: Microsoft Entra ID Application Proxy, F5 BIG-IP APM, NetScaler Gateway, NetIQ Access Gateway. SAP's own version is Network Edge Authentication with SAP Single Sign-On, where the Web Dispatcher performs the authentication step itself.
Identity runs on SAML 2.0. The front-end server is the Service Provider, the corporate IdP authenticates the user and applies MFA and conditional access, and the redirect the user sees is the SAML HTTP-Redirect binding. Identity then travels inward by principal propagation: a short-lived X.509 certificate, a header variable, or an SAP logon ticket.
This has two consequences. You can switch off form-based and basic authentication on the front-end server entirely, so there is no longer a password-guessing endpoint on the internet. And you get MFA at all, since SAP's own authentication stack has no second factor for the launchpad. With authentication at the IdP, SAP inherits whatever your identity team builds, without a separate SAP project.
Configuration that has to follow
The diagram only holds if each layer is configured to do its part. A few settings per layer decide that.
On the WAF, run the OWASP Core Rule Set in blocking mode rather than detection mode, rate-limit the authentication path, and check that you have virtual patching. When the next HotNews note comes out, you want to block a path within hours instead of waiting for a kernel window.
On the authentication proxy, allow no anonymous pass-through on any path, including health checks and static resources. Enforce MFA from external networks, set a shorter session lifetime than internally, and send every request to the SIEM.
The Web Dispatcher is where I find the most gaps. Configure wdisp/permission_table as a deny-by-default allowlist that permits the Fiori paths and explicitly denies the rest. Set icm/HTTP/support_http2 on purpose rather than leaving the default. Add HSTS, X-Content-Type-Options and a Content-Security-Policy with a frame-ancestors directive: a launchpad that can be framed can be clickjacked. Keep the kernel at or above the patch level of the systems behind it, and patch the host as an internet-facing asset, not as Basis infrastructure.
On the front-end server, go through SICF and deactivate every service outside the Fiori scope, starting with WebGUI, SOAP RFC, ADT and /sap/public/info. Set is/HTTP/show_detailed_errors = FALSE so stack traces stop leaking. Set icf/set_HTTPonly_flag_on_cookies and login/ticket_only_by_https. Activate HTTP security session management with a reasonable http/security_session_timeout. Then check that the active OData services are only those the published apps use; once a service is activated, nobody looks at it again. This is the same work as the rest of SAP platform hardening, on a host with no room for mistakes.
The back end should be reachable from the front-end server over trusted RFC with a restricted user, and from nothing else. The protocol change matters: OData in, RFC out, no HTTP path from edge to core. It is the best argument for hub deployment over embedded when the system faces the internet. If your Fiori server is embedded in S/4HANA, that decision is worth revisiting.
In the SIEM, you need both the ICM HTTP access logs and the ABAP security audit log, correlated with the authentication proxy logs. Without that correlation you cannot reconstruct what an authenticated session did once inside.
The BTP alternative
SAP Business Technology Platform avoids the exposure instead of mitigating it. SAP Build Work Zone hosts the launchpad, SAP Cloud Identity Services handles authentication, and SAP Cloud Connector links the on-premise system to the subaccount.
Cloud Connector builds its tunnel by reverse invoke, so the connection is opened outbound from inside your network. There is no inbound firewall rule, no published address and no listening service on the internet. The resource allowlist is enforced per path on your side of the tunnel.
It has costs: dependency on BTP availability, the subscription, and a Cloud Connector host that becomes a control point with its own hardening and monitoring. For organizations already on BTP it is usually the better option, and it is worth pricing before building the five-layer stack.
If you have inherited an exposed system
Do not patch first. Assume the window between exploitable and patched has already been used.
- Check ICM access logs for requests to paths no legitimate Fiori app calls.
- Look for files in the servlet and JSP directories that nobody created.
- Review users, role assignments and RFC destinations added since the last known-good state.
- Run a compromise assessment. Onapsis and Mandiant published a joint tool during the CVE-2025-31324 response; the approach applies beyond that CVE.
Then patch, then close the exposure. Patching first destroys evidence and does nothing about a web shell already on disk.
Wording for the audit file
External access is brokered by an authentication reverse proxy performing pre-authentication against the corporate identity provider, and the SAP Fiori launchpad is not directly reachable from untrusted networks.
The control objectives behind that sentence are standard. ISO 27001 Annex A requires network segregation and secure network services. NIS2 Article 21 requires network security and access control proportionate to the risk, and for entities in scope, an ERP running order-to-cash is part of that scope. An auditor with SAP experience will ask what sits between the internet and the ICM, and "the Web Dispatcher" will not satisfy them.
How MTC helps
From Geneva, we design and review SAP security architectures for regulated, multi-country landscapes. For external Fiori access this covers:
- reducing the ICF service tree
- Web Dispatcher and WAF configuration
- pre-authentication proxy and SAML design
- hub versus embedded deployment
- BTP and Cloud Connector alternatives
- the log correlation auditors ask for
On large programs we work alongside leading global audit, risk and technology consulting firms, with a senior Swiss-based team responsible for the security outcome.
Related reading
- SAP Security Consulting, our full SAP security scope
- SAP Cybersecurity: Hardening the SAP Platform, Not Just Access
- S/4HANA Migration: Greenfield vs Brownfield vs Selective
- IT General Controls (ITGC): A Practical Guide for Audit Readiness
- SAP GRC Access Control: What It Is and When You Need It
Frequently Asked Questions
Can you expose an SAP Fiori launchpad to the internet?
Not directly. A Fiori front-end server is a full ABAP application server, so publishing it puts the Internet Communication Manager and the whole Internet Communication Framework service tree within reach of unauthenticated attackers. External access should go through a pre-authentication reverse proxy that authenticates the user against the corporate identity provider before any request reaches SAP, with a web application firewall in front of it and the SAP Web Dispatcher behind it. The other option is SAP Build Work Zone on BTP with SAP Cloud Connector, which needs no inbound firewall rule.
Is the SAP Web Dispatcher a security boundary?
No. It terminates TLS, distributes load, hides internal hosts and ports, filters URLs against its permission table and gives the browser a single origin. It does not inspect payloads, rate-limit credential stuffing, enforce MFA or decide who the user is; by default it forwards. It also runs the same ICM code as the application servers behind it, which is why CVE-2022-22536 affected it too. It has a place in the architecture as a reverse proxy, not as the only thing between the internet and the ABAP stack.
What is an authentication proxy for SAP Fiori?
A reverse proxy that authenticates the user against the corporate identity provider before forwarding anything to SAP, so no anonymous request reaches the ABAP stack. Vendors sell it as an identity-aware proxy, access gateway, web access management or ZTNA broker: Microsoft Entra ID Application Proxy, F5 BIG-IP APM, NetScaler Gateway, NetIQ Access Gateway. SAP's own variant is Network Edge Authentication with SAP Single Sign-On, where the Web Dispatcher performs the authentication step. Identity runs on SAML 2.0, with the front-end server as Service Provider and principal propagation carrying the identity inward.
Which ICF services should be deactivated on an internet-facing Fiori front-end server?
Fiori needs about a dozen paths: the launchpad shell under /sap/bc/ui2/, UI5 resources under /sap/bc/ui5_ui5/ and OData services under /sap/opu/odata/. Review the rest of the SICF tree and switch it off, starting with /sap/bc/gui/sap/its/webgui (SAP GUI in a browser), /sap/bc/soap/rfc (RFC over SOAP), /sap/bc/webdynpro/, /sap/bc/adt (the ABAP development endpoint) and /sap/public/info, which returns the SID, kernel release, host name and time zone without authentication. A default S/4HANA front-end server has hundreds of ICF nodes active.
What did CVE-2025-31324 change about SAP internet exposure?
It showed how long the gap between exploitable and patchable can be. The flaw was a missing authorization check in the NetWeaver Visual Composer metadata uploader (CVSS 10.0): an unauthenticated POST to one endpoint uploaded a JSP web shell. Unit 42 saw probing in late January 2025, Rapid7 confirmed exploitation from at least 27 March, and SAP issued an out-of-band patch on 24 April. For a month, arguably three, any internet-facing system with the component enabled could be compromised with no patch, no signature and no advisory. Patching could not help during that period; not being reachable could.
Is SAP BTP a safer way to publish Fiori externally?
Usually, because it avoids the exposure instead of mitigating it. SAP Build Work Zone hosts the launchpad, SAP Cloud Identity Services handles authentication, and SAP Cloud Connector links the on-premise system to the subaccount by reverse invoke, so the tunnel is opened outbound from your network. There is no inbound firewall rule, no published address and no listening service on the internet, and the resource allowlist is enforced on your side of the tunnel. The costs are dependency on BTP availability, the subscription, and a Cloud Connector host that needs its own hardening and monitoring.