SAP's November patches: four flaws that need no password
SAP fixed four flaws rated 10.0 that need no password, in the web front door, the Web Dispatcher, the gateway and SAML single sign-on. Patch the kernel and lock down the gateway's proxy list this week.
part 41 of 41 · Platform releases, explained from scratchthe ideaFour no-login doors into SAP: close the gateway list, then patch the edge first.
after reading, you can- Rank November's SAP fixes so internet-facing Web Dispatchers and gateways get patched first
- Check each kernel against the fixed levels, and watch for the 7.93 patch level 419 trap
- Close the gateway proxy list and set gw/acl_mode_proxy so the fix actually holds
SAP fixed four flaws this month that let a stranger take over an SAP system without a password. They sit in the parts that face the network: the web front door, the gateway other systems talk to, and single sign-on. Patch the kernel and every Web Dispatcher now, and lock down the gateway's proxy list before you wait for a maintenance window.
If you run SAP on your own servers or in a private cloud, don't leave this month's fixes for the next window.
SAP systems run the money side of a business: orders, invoices, payroll, stock. Every one of them has a few built-in pieces that talk to the outside world. One answers web traffic, one is the reverse proxy many companies put in front of it, one lets other programs connect, and one handles single sign-on. This month SAP fixed holes in all four that someone could use with no account at all. The worst outcome is the same for each: the attacker controls the system and everything it can reach. Nobody has reported attacks yet, but SAP bugs this serious have been attacked within days before.
Think of it as an office tower where the lobby door could be jammed open. The loading dock waved through anyone who said they were staff, and the badge reader accepted a badge nobody had ever issued.
- Companies running S/4HANA or ECC
- Anyone with Fiori open to the internet
- Teams using SAML single sign-on into SAP
- RISE with SAP customers
- SAP Basis and security teams
Why it matters to you
An SAP system holds the ledger, the vendor bank details and the customer master. Whoever controls it can read or change any of that, quietly. Researchers at Onapsis, who found three of the four 10.0 flaws, say their exploits run commands as <sid>adm, the operating system account that owns the SAP installation. From there the database password and the stored connections to other systems are in reach.
Nothing from this month is known to be under attack. CISA's catalog of exploited bugs had none of these as of November 13, and Onapsis says it hasn't seen attacks. That's the good news, and it may not last. In August a critical SAP Commerce Cloud flaw drew attacks within days of its fix, according to threat-intelligence firm Defused, and SAP already has 14 entries on CISA's list.
Three of the four bugs also need nothing unusual switched on. WebSocket support is enabled by default in the web component and the Web DispatcherdefinitionSAP's own reverse proxy. Many companies put it at the network edge so Fiori and API traffic goes through it before reaching the SAP system., and the gateway's proxy setting ships in allow-all mode for backward compatibility. A system nobody has touched in years can be exposed exactly as it was installed.
-
01 Attacker reaches the gateway
Each ABAP server has a gateway listening on port 33NN or 48NN. Anyone whose network reaches that port can have a go.
-
02 The proxy list lets it through
The gateway checks a proxy list called prxyinfo first. Where nobody wrote that file and the default was left alone, it waves every proxy request through.
-
03 It now looks like an inside server
A few crafted requests later, the gateway thinks it's talking to one of the system's own application servers.
-
04 The usual rules say yes
Most hardened setups only trust the system's own servers. The gateway now thinks this request came from one of them, so the rules wave it through.
-
05 Commands run as <sid>adm
The gateway starts what the attacker asked for. In the worst case that's any operating system command, run as the account that owns SAP.
The awkward part for Basis teams is where the hole sits. Most gateway hardening guidance covers two lists, secinfo and reginfo, and Onapsis says neither one stops this attack. The list that matters is prxyinfodefinitionThe SAP Gateway's proxy access list. It decides which connections the gateway may relay or open for a client. If it's missing, the gw/acl_mode_proxy setting decides., the one fewest reviews ever open. A gateway that passed last year's audit can still be wide open here.
What SAP fixed, and the catches in each fix
The four 10.0s first, then the five other Critical notes you'll still need
Note 3778530. Crafted WebSocket messages overflow memory in the ICM, the piece that handles the server's web traffic, and nobody has to log in. AS Java is safe, but dual-stack systems aren't.
Note 3778556. Same root cause in the reverse proxy, including the Web Dispatcher built into SAP HANA XS. One forgotten instance keeps the hole open.
Note 3798345. The gateway's default proxy settings let an outside attacker pass as an internal server. The fix depends on a profile parameter as well as the kernel patch.
Note 3849308. Under certain configurations, AS ABAP accepts a sign-in assertion without checking it properly. An attacker could log in as anyone, administrators included. RedRays, which reviewed the note, says there's no workaround.
OS commands through SAPFTP (3792735), code injection in S4CRM and SAP CRM through an unchecked remote function (3834842), and portal XML Forms (3809153). One phished login is enough.
Note 3804858 (9.6) needs a logged-in ICM or Web Dispatcher admin to click a link. Note 3799117 (9.1) lets a BusinessObjects user read server files and reach internal systems.
Versions before 0.63.1 read the tenant and user from a JWT without checking its signature, so a caller could pose as another tenant in logs, audit records and tenant routing.
The SAML2definitionSingle sign-on for SAP, usually through Entra ID or Okta. The identity provider signs each sign-in, and SAP is meant to check that signature every time. flaw is the one identity teams should read twice. Single sign-on moves trust from the password to the identity provider's signature on each sign-in. If the system doesn't check that signature, anyone who can reach the sign-in endpoint can write their own. SAP's own CVE description says the attacker can impersonate arbitrary users, privileged accounts included, and the damage can spread to connected systems.
SAP lists 16 affected SAP_BASIS releases, from 702 up to 922, so age won't get a system off the list. The trigger is configuration. RedRays, a firm that reviewed the gated notes, treats any ABAP system that uses SAML2 for browser sign-in as in scope until someone proves otherwise, and plenty of Fiori systems do. It also gives a quick test: if the system answers a request for its SAML2 service-provider metadata, SAML2 is on for that client.
ICM2SHELL, Onapsis's name for the WebSocket bug, has two traps. On kernel 7.93, a higher patch level isn't always a fixed one: patch level 419 is still vulnerable, while the updated 401 stack kernel and 420 or later carry the fix. Kernel releases that are out of maintenance, such as 7.49, 7.73 and 7.89, are affected too and may not get a fix, so those systems need a kernel upgrade.
PROXYGATE has the subtlest fix. On kernels 7.93 and 9.16 the setting already exists, and the fix is to set gw/acl_mode_proxy to 1. Kernels 7.22, 7.53, 7.54, 7.77 and 8.04 only gain that setting with the November kernel patch. Onapsis is blunt about it: installing the patch without a restrictive prxyinfo file or the parameter set to 1 leaves the system in scope.
| Kernel release | ICM and Web Dispatcher (ICM2SHELL) | Gateway (PROXYGATE) | Extra step |
|---|---|---|---|
| 7.22 EXT / EX2 / EX3 | Not affected | PL 1612 | Set gw/acl_mode_proxy = 1 after patching |
| 7.53 | PL 1616 or 1617 | PL 1625 | Set gw/acl_mode_proxy = 1 after patching |
| 7.54 | PL 710 | PL 718 | Set gw/acl_mode_proxy = 1 after patching |
| 7.77 | PL 914 | PL 915 | Set gw/acl_mode_proxy = 1 after patching |
| 7.93 | Updated PL 401 stack kernel, or PL 420 and later (PL 419 is still vulnerable) | Parameter already exists | Set gw/acl_mode_proxy = 1 now |
| 8.04 | Not in SAP's affected list for this bug | PL 244 | Set gw/acl_mode_proxy = 1 after patching |
| 9.16 | PL 111 | Parameter already exists | Set gw/acl_mode_proxy = 1 now |
| 9.18, 9.20, 9.21 | 9.18 PL 37; 9.20 PL 11 or 13; 9.21 not listed | 9.20 PL 16, 9.21 PL 4 | Apply the kernel patch |
| Out of maintenance: 7.40, 7.41, 7.42, 7.45, 7.49, 7.73, 7.81, 7.85, 7.89 (and 7.20, 7.21 for the gateway) | Affected, may get no fix | Affected, may get no fix | Move to a supported downward-compatible kernel |
| SAP HANA 2.0 (Web Dispatcher in XS Classic) | Revisions 2.00.079.11 or 2.00.089.05 | n/a | Patch HANA as well as the kernel |
| XS Advanced (XS Runtime 1) | SP05 Patch 4 | n/a | Patch XS Advanced too |
- No prxyinfo file, so the gateway falls back to its default
- gw/acl_mode_proxy still at 0, so any proxy connection gets in
- secinfo and reginfo hardened, which doesn't help here
- Gateway ports reachable from user networks
- Still exploitable, despite the green patch report
- A restrictive prxyinfo naming only trusted hosts
- gw/acl_mode_proxy = 1, so a missing file fails closed
- Kernel at or above the fixed patch level
- Ports 33NN and 48NN closed to everyone who doesn't need RFC
- Out of scope even if someone deletes the file later
How the month has gone so far
- Oct 29, 2026Out-of-band note for Node.js on BTP
SAP published note 3843368 for the Brotli decoder flaw in Node.js buildpacks. Running Cloud Foundry apps only pick up the fix after a restage.
- Nov 5, 2026Out-of-band note for PHP on BTP
Note 3846599 covers the PHP2Shell chain. SAP rated both of these out-of-band notes 9.8, and both went out before Patch Day.
- Nov 10, 2026Patch Day
SAP put out 35 new notes, 4 GitHub advisories and 3 updated notes. Onapsis posted its ICM2SHELL and PROXYGATE write-ups that same day.
- Nov 13, 2026Still not on the KEV list
CISA's catalog version 2026.11.13 lists none of the November SAP flaws. That's a status to recheck every week, not a promise.
- Dec 8, 2026Next SAP Patch Day
SAP's next regular Patch Day is the second Tuesday of December. Any kernel still unpatched by then turns into a bigger job.
What to do now
Most of this month's work is one kernel rollout. RedRays counts 17 of the November fixes in the kernel, the standalone Web Dispatcher or both, so the sensible move is a single update that covers them all instead of one patch per note. The SAML2 fix is the exception: it's an ABAP correction, delivered through the note's correction instructions or a support package.
Who does what this week
- Ask your SAP Basis team or provider for a written date for the November kernel and SAML2 fixes on production.
- If you're on RISE with SAP, ask your SAP contact when the notes land in your landscape. SAP applies kernel patches on its own schedule.
- Ask whether any SAP web address or gateway port can be reached from the internet. If nobody knows, that's the first job.
- Tell your insurer and lawyer early if anyone finds signs of a break-in. Notice deadlines can be short.
- List every kernel, every Web Dispatcher and every HANA XS instance, including the small ones in front of HANA or third-party systems.
- Patch internet-facing Web Dispatchers first, then production application servers, then the rest.
- Check prxyinfo on every system. Restrict it now, then set gw/acl_mode_proxy = 1 wherever the kernel supports it.
- Apply note 3849308 on every system where any client uses SAML2 sign-in, then confirm each server's kernel level in transaction SM51.
- Open transaction SAML2 on each ABAP system and review every trusted identity provider. Remove the stale ones.
- Pull security audit log entries for SAML sign-ins by privileged users and look for sources you can't explain.
- Ask developers which apps read IAS tokens through sap-cloud-sdk for Python, and move them to 0.63.1 or later.
- Scan from outside your network for exposed ICM, Web Dispatcher and gateway ports. Don't rely on the asset list.
- Watch dev_icm and dev_webdisp for unexplained restarts. Onapsis expects each exploit attempt to crash the process at least once.
- Alert on new processes started by <sid>adm and on gateway log entries for programs that never ran before.
# Does this client answer as a SAML2 service provider? XML back means SAML2 is on $ curl -s "https://<host>:<port>/sap/saml2/sp/metadata?sap-client=100" | head -c 200 # Look up where the proxy list lives (profile parameter gw/prxy_info), then open it $ cat <path from gw/prxy_info> P SOURCE=* DEST=* # That wildcard line means exposed at any patch level
A sensible order for the rollout
- 01Build the list
Record each system's kernel release and patch level, its Web Dispatchers, whether it uses SAML2, the state of prxyinfo and the value of gw/acl_mode_proxy. Include sandboxes holding copies of production data.
- 02Close the proxy list today
Restrict any permissive prxyinfo to trusted hosts, starting with gateways reachable from outside. Test it in a non-production system first, because some gateway-to-gateway setups depend on proxy rules.
- 03Patch the edge
Update internet-facing Web Dispatchers first. Onapsis notes that a Web Dispatcher at 7.53 PL 1629, 7.54 PL 723, 7.93 PL 430, 9.16 PL 200 or higher can shield backends reached only through it, but not one set up as a router.
- 04Roll the kernel through
Roll every application server to a kernel that covers all of November's kernel notes and set gw/acl_mode_proxy = 1 afterwards. Out-of-maintenance kernels may not get a fix. Move those to a supported downward-compatible kernel.
- 05Apply the SAML2 correction
Implement note 3849308 on every SAML2 system, production first. There's no workaround, so a system waiting for a transport window stays exposed.
- 06Prove it
Confirm patch levels after the restarts, recheck prxyinfo and the parameter, and close the change with the list of systems and levels attached.
Harden so the next one hurts less
All four 10.0s sit in components that answer network traffic, and often more networks can reach them than need to. This month 17 of the fixes sit in the kernel or the Web Dispatcher alone. Shrinking who can reach those pieces helps every month, whatever the next bug turns out to be.
Before: open edge, open gateway
The Web Dispatcher sits on the internet unpatched. User networks can reach the gateway, and nobody has looked at the SAML2 trust list in a while.
After: patched edge, closed gateway
The edge is patched, gateway ports only answer known RFC partners, and a missing proxy file now fails closed.
Durable fixes, tied to this month
- 01Keep gateways off user networks
Allow ports 33NN and 48NN only from the application servers and RFC partners that need them. An SAP Gateway has no reason to be reachable from the internet.
- 02Make missing files fail closed
Set gw/acl_mode_proxy = 1 and keep a reviewed prxyinfo, secinfo and reginfo on every system. Add all three files to the yearly audit.
- 03Put one patched door in front
Route Fiori and API traffic through a Web Dispatcher you patch first, and make sure nothing can reach the ICM ports around it. Prove that with a scan.
- 04Review SAML2 trust like a password list
Each trusted identity provider in transaction SAML2 can sign anyone in. Review the list every quarter, and alert on privileged SAML sign-ins from unusual sources.
- 05Patch the kernel every quarter at least
November alone brought 17 kernel and Web Dispatcher fixes. A kernel that only moves during upgrade projects will usually be months behind.
- 06Lock up what <sid>adm can see
Keep the secure store tight, rotate the database and RFC technical passwords, and limit outbound internet access from SAP hosts so a foothold can't phone home.
If you've been breached
If you think someone used one of these flaws
Warning signs include ICM or Web Dispatcher crashes you can't explain, new processes run by <sid>adm, gateway log entries for unfamiliar programs, and privileged SAML sign-ins from odd sources. Patching won't remove anyone already inside.
- 01Containfirst hour
- Block the attacker's route in: take the Web Dispatcher off the internet or close the gateway ports at the firewall.
- Keep the servers running so memory and work-directory logs survive for the investigators.
- Lock the SAP accounts that signed in through SAML from unknown sources, and disable the suspect identity provider trust.
- 02Investigatefirst day
- Collect dev_icm, dev_webdisp, the gateway log and the security audit log before they rotate.
- Look on the host for files, cron jobs and services that <sid>adm created in recent weeks, especially since November 10.
- Check for new dialog users, changed authorizations and new RFC destinations in the ABAP system.
- 03Removefirst days
- Rebuild a compromised application server or Web Dispatcher host from known-good media, then patch before it goes back on the network.
- Delete accounts, roles and RFC destinations the attacker created, and record each one.
- 04Recoverthis week
- Rotate the database password, the secure store contents and the technical passwords in RFC destinations, because <sid>adm could read them.
- Check connected systems that trusted this one, such as other SAP systems over trusted RFC, for the same signs.
- Reconcile recent payment runs and vendor bank detail changes with finance before the next run.
- 05Report and learnas required
- Work with legal on notice duties. Customer, employee or financial data in SAP can trigger regulator and customer notifications.
- Write down how the attacker got in and which control would have stopped it, then fund that control.
What this could look like: three typical setups
Three typical SAP estates and November's fixes
- the setup
- A mid-size manufacturer runs S/4HANA on its own servers. Sales reps use Fiori from the road, so a Web Dispatcher faces the internet. The kernel was last patched in spring.
- what happens
- The Web Dispatcher still has WebSocket on, because nobody knew it was enabled by default. Anyone on the internet can reach the vulnerable code without a login, and a crash takes Fiori down for every rep.
- what to do
- Patch the Web Dispatcher this week, ahead of the application servers. Scan from outside to confirm nothing reaches the ICM ports around it, then roll the kernel through.
- the setup
- A distributor's auditors signed off on hardened secinfo and reginfo files last year. The ECC system runs kernel 7.53, there's no prxyinfo file, and the office network can reach the gateway port.
- what happens
- None of the audited rules stop PROXYGATE. With no proxy file and the parameter not available on that kernel, the gateway allows every proxy connection, so anyone on the office network could pass as an internal server.
- what to do
- Create a restrictive prxyinfo today. Patch to 7.53 PL 1625 or later, set gw/acl_mode_proxy = 1, and close port 33NN to everything except the RFC partners that need it.
- the setup
- A retailer moved to RISE with SAP. Staff sign in to Fiori through Microsoft Entra ID over SAML2. SAP runs the kernel, while the retailer owns network access and identity settings.
- what happens
- The kernel fixes arrive on SAP's maintenance schedule, which may be weeks away. Until then the retailer's exposure depends on what it lets reach the system, and the SAML2 trust list still includes a test identity provider from the migration.
- what to do
- Ask SAP for the date the November notes reach each system. Meanwhile, remove the stale identity provider, review privileged SAML sign-ins and limit which networks reach Fiori and the gateway.
How Cohash helps
Our SAP Basis and security engineers can build the kernel, Web Dispatcher and gateway inventory with your team, plan one rollout that covers the November notes and review the SAML2 trust lists with your identity team. The SAP S/4HANA page explains how we work on S/4HANA and ECC, you can bring in SAP engineers for the patch window, or start with an assessment of what your landscape exposes. Two of SAP's out-of-band notes since October's Patch Day cover flaws we've already explained: the Brotli decoder bug and the PHP and Node.js releases.
What to take from this
- 01
Four of November's SAP flaws need no login. Patch the kernel and every Web Dispatcher this week, edge first.
- 02
On most kernels the patch alone won't fix the gateway flaw. Restrict prxyinfo and set gw/acl_mode_proxy = 1.
- 03
On kernel 7.93, patch level 419 is still vulnerable. Use the updated 401 stack kernel or 420 and later.
- 04
Apply the SAML2 correction on every system that uses SAML sign-in, and review the trusted identity providers.
- 05
None of this is under attack yet. SAP flaws have been attacked within days before, so don't wait for the KEV list.
- How Cohash delivers SAP S/4HANA: scope, roles and architecture
- How Cohash delivers SAP EWM, TM & IBP: scope, roles and architecture
- How Cohash delivers SAP IS-U, BRIM & FI-CA: scope, roles and architecture
- Hire vetted SAP S/4HANA engineers
- ERP systems
- Book a 30-minute working session with a senior engineer
01Does the November patch day affect our SAP system?
Almost certainly, if you run NetWeaver ABAP, S/4HANA, ECC or a standalone Web Dispatcher yourself or in a private cloud. Every ABAP system runs a kernel, and that's where most of the critical fixes are. Check the kernel release and patch level first, then whether anyone signs in through SAML2.
02Are any of these SAP flaws being exploited?
Not as of November 13. CISA's exploited-vulnerabilities list doesn't include them, and Onapsis, which found three of the four 10.0 flaws, says it hasn't seen attacks. SAP flaws have been attacked within days of a fix before, so plan as if that will change.
03We hardened secinfo and reginfo. Are we safe from the gateway flaw?
No. Onapsis says those two lists play no part in stopping PROXYGATE. What matters is the proxy list, prxyinfo, and the gw/acl_mode_proxy setting when that file is missing. Restrict the proxy list, then patch.
04We put the November kernel patch in. Is the gateway flaw closed now?
On most kernels, not on its own. After patching, the gateway still needs a restrictive prxyinfo file or gw/acl_mode_proxy set to 1. On 7.93 and 9.16 the setting already exists, so you can set it today. On 9.20 and 9.21 the patch alone is SAP's fix.
05We're on RISE with SAP. Is any of this our job?
Some of it is. SAP patches the kernel on its own maintenance schedule, so ask your SAP contact for the date each system gets the November notes. What can reach the system, and which identity providers it trusts, is still yours to manage. SAP note 3849055 is the private cloud FAQ for these notes.
06Can we turn off WebSocket instead of patching?
Only as a stopgap. Setting icm/HTTP/support_websocket_upgrade to FALSE blocks the ICM2SHELL bug, but SAP warns it can affect ABAP Push Channels and some Fiori features. Check the HTTP log for status 101 requests before you do it, and patch anyway.
07Does the SAML2 flaw affect systems that don't use single sign-on?
SAP only says the flaw shows up under certain configurations. The people who've read the note focus on clients that use SAML2 for browser sign-in. Try the client's SAML2 metadata address: if it answers, SAML2 is on there and note 3849308 goes in.
Three questions, about a minute
Pick an answer and we'll tell you if it's right, and why it is or isn't. The score stays in this browser.
-
Q1Your ECC system is on kernel 7.53. Someone hardened secinfo and reginfo, but nobody ever created a prxyinfo file. What keeps PROXYGATE out?
show the answer
B. Close off prxyinfo now. Patch next, and set the parameter to 1 Secinfo and reginfo don't stop this attack. With no proxy file, the gateway allows every proxy connection until the patched kernel lets you set gw/acl_mode_proxy to 1, so a restrictive prxyinfo is the interim control.
-
Q2A Basis admin reports every server is on kernel 7.93 patch level 419, higher than the 401 listed in the note. Is ICM2SHELL closed?
show the answer
D. No, 419 is still vulnerable; use the updated 401 stack kernel or 420 and later On 7.93 a higher patch level isn't always a fixed one. Onapsis, citing SAP's FAQ note 3778788, lists 419 as still affected, so those servers need 420 or later, or the updated 401 stack kernel.
-
Q3Your Fiori system signs users in through SAML2 and the correction can't go in until next month's transport window. What's the honest status?
show the answer
A. Exposed until the correction is in, because there's no workaround The flaw lets a forged sign-in through without a valid signature, so the identity provider's own checks don't help. RedRays, which reviewed the note, says there's no workaround. Reviewing trusted identity providers and privileged SAML sign-ins limits the damage, but only the correction closes it.