9.8

CVSS 3.1 score Oracle assigned to CVE-2026-101767

7 days

from Oracle's patch to the first probing Defused reported

31

EBS flaws in the October CPU exploitable without credentials

4

EBS vulnerabilities already in CISA's exploited catalog

The patch date is now the attack date

Oracle published the fix for CVE-2026-101767 on 20 Oct 2026. Seven days later, someone was already sending exploit attempts at E-Business Suite systems on the internet. That gap is the story, and it is the reason this post exists.

The bug is serious on its own terms: an unauthenticated SQL injection in the Application Object Library, the core of E-Business Suite, rated 9.8, reachable over HTTP, with no user interaction. The researchers who found it say it ends in operating system commands on the application server. But plenty of 9.8s come and go. What makes this one worth a long read is the collision between two calendars.

The first calendar belongs to the people who run EBS. It is the system that pays suppliers, closes the books and holds payroll and procurement data. Patching it means downtime, regression testing of customizations, and a change board that wants to know why finance cannot post journals on a Tuesday. Oracle ships its Critical Patch Updates quarterly, and EBS patching tends to follow a planned window after each one, moving through test before it reaches production. That rhythm made sense when the CPU was the start of a slow, quiet period.

The second calendar belongs to attackers, and it starts on patch day. A patch is a map. Researchers and criminals both diff it, and for EBS the payoff for being early is proven: in 2025 an extortion crew used the Cl0p brand to steal data from EBS environments through a zero-day, and the attention that campaign drew has not gone away. A week from patch to probing is not surprising. It is what a mature target looks like.

Our argument is simple. If your EBS patching is scheduled around the next maintenance window, it is scheduled around the wrong date. The October CPU is the fix. The control you can ship tonight, without touching the application, is the URL firewall allowlist, which in the default configuration lets the vulnerable servlet through. Close that, keep the web tier off the open internet, and you have bought the time your change board needs.

What is CVE-2026-101767 in Oracle E-Business Suite?

Oracle's October 2026 Critical Patch Update, released on 20 Oct, contains 1295 new security patches across its product families. One hundred and ninety four of them are for E-Business Suite, and Oracle says 31 of those "may be remotely exploitable without authentication." CVE-2026-101767 is near the top of that list.

The NVD record, which carries Oracle's own text, describes an "easily exploitable vulnerability" in the Oracle Application Object Library, component Core, that "allows unauthenticated attacker with network access via HTTP to compromise Oracle Application Object Library." Successful attacks "can result in takeover" of it. The supported versions affected are 12.2.3 through 12.2.15. The CVSS 3.1 base score is 9.8, with the vector every defender dreads: network attack vector, low complexity, no privileges, no user interaction, high impact on confidentiality, integrity and availability.

Horizon3.ai found and reported it, and its write up adds the detail Oracle's matrix leaves out. The flaw sits behind the TCF servlet at /OA_HTML/AppsTCFServer. Horizon3 says the attacker can "execute SQL as the APPS database user, then run operating system commands as the EBS application account." It also says the one thing that should move this from the patch queue to tonight's agenda: the endpoint "is permitted through the default EBS URL firewall configuration." Horizon3 reported the flaw to Oracle on 24 Jul 2026, which put 88 days between report and fix.

CVE-2026-101767 Critical
  • Product //

    Oracle E-Business Suite, Application Object Library, Core
  • Affected //

    12.2.3 through 12.2.15
  • Fixed in //

    October 2026 Critical Patch Update, 20 Oct 2026
  • CVSS //

    9.8 (3.1, Oracle)
  • Access //

    HTTP, no authentication, no user interaction
  • Exploited //

    Probing reported 27 Oct; not in CISA KEV as of 27 Oct

The APPS account matters more than the SQL injection label suggests. In EBS, APPS is the schema the application itself runs as. It can read the general ledger, supplier bank details, employee records and the application's own configuration. SQL as APPS is not a foothold in the database. It is the database, as far as the business data is concerned. The step from there to operating system commands on the application tier is what turns a data breach into a host compromise, and it is the step that gives an attacker persistence after the patch goes in.

Is CVE-2026-101767 being exploited?

Here we hedge exactly as much as the sources do.

On 27 Oct the threat intelligence firm Defused reported that its EBS decoys had "recorded exploratory exploits" against CVE-2026-101767 that morning. SecurityWeek carried the report and added that "there have been no other reports regarding its in-the-wild exploitation." So the honest statement is: attackers are probing for the bug on systems built to look like EBS. Nobody has publicly confirmed a real organization compromised through it.

CISA's own view, recorded on the NVD entry on 21 Oct, the day after the patch, rates exploitation "none," automatable "yes" and technical impact "total." The first of those was true when it was written and is now out of date. The second and third are the ones to weigh. Automatable means an attacker can script it across every exposed target without hand work. Total means full control of the component once it lands. CVE-2026-101767 was not in CISA's Known Exploited Vulnerabilities catalog in the release published on 27 Oct.

The history explains why probing is a warning rather than noise. E-Business Suite already has four entries in the KEV catalog: CVE-2022-21587, added in February 2023; CVE-2025-61882 and CVE-2025-61884, added in October 2025; and CVE-2026-46817, added on 15 Jul 2026. CISA marks the first three as known to be used in ransomware campaigns.

CVE-2025-61882 is the one everyone remembers. According to Google Threat Intelligence Group, it was exploited as a zero-day as early as 9 Aug 2025, and extortion emails claiming affiliation with the Cl0p brand began reaching executives on 29 Sep 2025. Oracle's emergency patch followed on 4 Oct. That is 56 days of exploitation before any fix existed. SecurityWeek reports that the campaign potentially affected more than 100 organizations. GTIG did not formally attribute the activity to a tracked group, and noted overlaps with activity associated with FIN11.

So the pattern is established. EBS sits on the internet more often than it should, it holds data worth extorting, and at least one well resourced crew has already built tooling around its web tier. A new unauthenticated 9.8 in the same web tier, published with a researcher write up naming the endpoint, will not stay at the probing stage for long. We would plan as if it will move to confirmed exploitation within weeks, and be glad if it does not.

How an unauthenticated request becomes an EBS takeover

We describe the chain at the level of what each stage achieves, which is the level a defender needs. Horizon3 has published no payload and neither do we.

Figure 01

From one web request to the EBS host

  1. 01

    Reach //

    An attacker reaches the EBS web tier and the TCF servlet the URL firewall allows.

  2. 02

    Inject //

    One unauthenticated request runs SQL as the APPS user.

  3. 03

    Execute //

    Database access becomes commands as the EBS application account.

  4. 04

    Stay //

    Business data leaves, and a foothold survives the patch.

Result //

No credential is ever needed, and step 01 is the only one you control tonight.

Steps 02 and 03 are the bug. Step 01 is configuration, which is why the URL firewall matters before the patch lands.

Two things about this chain deserve a defender's attention.

First, step 01 is a choice. EBS deployments that serve suppliers or remote staff often put part of the web tier in a DMZ, and Oracle's URL firewall is the list of pages allowed through that external tier. Horizon3's point is that the default list includes /OA_HTML/AppsTCFServer. Removing that entry is the interim mitigation Horizon3 recommends, alongside restricting the web tier to trusted networks. Neither replaces the patch. Both are changes to a configuration file and a network rule, not to the application, which is a much easier conversation with a change board.

Second, step 04 is why patching late is worse than patching slowly. In the 2025 campaign GTIG found implants that lived mostly in memory inside the EBS Java processes, and it recommended memory forensics for suspected compromise. A patch applied after an intrusion closes the door behind the attacker. It does not evict them. If your EBS web tier was reachable from the internet at any point after 20 Oct, the question is not only whether you patched but whether anything happened in between.

How to detect CVE-2026-101767 attempts

Horizon3 lists four indicators, and they map cleanly onto the logs most EBS shops already have.

The web tier signal is an unauthenticated POST to the TCF servlet that instantiates the Gantt data server class, oracle.apps.jtf.gantt.server.GanttDataServer, followed by database activity. You will not see the class name in an access log, but you will see the POST, the path and the source address. On many deployments there is no business reason for that path to receive requests from the internet at all, which makes any external hit worth a look.

The database signal is reconnaissance SQL arriving soon after unauthenticated web requests: queries for the database version, the current user and directory objects. If you have database auditing on the APPS schema, that pattern is distinct from normal application traffic, which does not ask the database who it is.

The host signals are unexpected files in $APPLPTMP, an EBS temporary directory, and the application account spawning /bin/sh. Horizon3 names the oracle account and the Concurrent Manager path. GTIG's 2025 guidance named Java processes under applmgr starting interactive shells. Account names vary by installation, so baseline yours first.

The rule below covers the web tier, which is the signal most teams can deploy today.

Figure 02

POST to the EBS TCF servlet

Sigma

title: Oracle EBS AppsTCFServer POST request
status: experimental
description: A POST request to the Oracle E-Business Suite TCF servlet, the endpoint behind CVE-2026-101767. External requests to this path are rarely legitimate and warrant review.
references:
  - https://horizon3.ai/attack-research/vulnerabilities/cve-2026-101767/
  - https://nvd.nist.gov/vuln/detail/CVE-2026-101767
logsource:
  category: webserver
detection:
  selection:
    cs-method: 'POST'
    cs-uri-stem|contains: '/OA_HTML/AppsTCFServer'
  condition: selection
falsepositives:
  - Internal EBS features that legitimately use TCF, from trusted networks
  - Authorized vulnerability scanning
level: high
tags:
  - attack.initial_access
  - attack.t1190
  - cve.2026-101767
A starting point to tune against your own telemetry, not a production rule. Scope it to requests from outside your trusted networks, and pair it with an alert on the EBS application account starting a shell.

Run it backwards as well as forwards. Search the web tier logs from 20 Oct onward for the path, and if you find external POSTs, treat the host as a hunt, not a patch ticket. GTIG's 2025 advice still holds for that hunt: block outbound internet traffic from EBS servers that do not need it, so an implant cannot reach its controller even if one landed.

How we test EBS exposure on an engagement

On a sanctioned external test, an ERP web tier is one of the first things we look for, because it is one of the few systems where a single bug reaches money and personal data at the same time. The work is less exotic than the CVE suggests.

We start with discovery. Our external recon maps every host name and address the client owns, from DNS, certificate transparency logs and the ranges in the scope letter, and EBS has recognizable paths and login pages. The external recon page in our Learn section walks through that method, and our open source atarus-recon tool automates the first pass. The finding is often not a vulnerability at all. It is that a system the client believed was internal answers from the internet.

Then we check what the URL firewall actually allows, from the outside, against what the client expects it to allow. The allowlist is a file someone edited years ago, and no one has read it since. Comparing it to the pages the business really needs from outside is a short exercise with a large payoff, and it applies to every future EBS bug, not just this one.

We confirm exposure without exploitation. For a bug like CVE-2026-101767, version and patch level evidence plus reachability of the servlet is enough to rate the finding. We do not fire SQL injection at a production general ledger to prove a point, and a client should be wary of any tester who offers to. Where the client wants proof of impact, we agree a test instance and a written plan first.

Finally, we look at what happens after the foothold, because that is where most of the damage in 2025 happened. Can the application tier reach the internet? Is anyone alerted when the application account starts a shell? Would anyone notice a new file in the temporary directory? Those questions belong in any post exploitation review, and our web app persistence audit covers the application side of the same idea.

What is still coming for EBS

CVE-2026-101767 is the bug with the write up, which is why it is the one being probed first. It is not the only reason to take the October CPU seriously.

The same E-Business Suite risk matrix lists CVE-2026-101809, in the Workflow Notification Mailer, at CVSS 10.0: remotely exploitable without authentication over HTTP, low complexity, with a changed scope, across the same 12.2.3 to 12.2.15 range. No public write up exists for it yet. When one appears, expect the same week long gap to probing. Patching CVE-2026-101767 alone, by hand, on the theory that it is the dangerous one, leaves that door and 29 other unauthenticated ones open.

Oracle's own CPU text makes the point more bluntly than we would. It says it continues to receive reports of attempts to exploit flaws it has already patched, and that in some instances "attackers have been successful because targeted customers had failed to apply available Oracle patches." That sentence has appeared in Oracle advisories for years. It describes exactly the gap between the two calendars.

We also expect more research on the EBS web tier, not less. Two campaigns in two years have shown that it holds data worth stealing and is reachable often enough to be worth scanning. Each new public write up shortens the next attacker's work. The structural fix is to make the web tier unreachable from places it does not need to be reached, so that the next CPU is a maintenance task rather than a race. We made the same argument about a different tier zero system in our VeloCloud Orchestrator analysis: patching each advisory is necessary, and isolating the system is what stops the next one from becoming an emergency.

What to do today

  1. Apply the October 2026 CPU to every EBS 12.2.3 to 12.2.15 instance. Use Oracle's E-Business Suite CPU knowledge document, and take the whole update, which also fixes the 10.0 in the Workflow Notification Mailer.

  2. Remove /OA_HTML/AppsTCFServer from the URL firewall allowlist tonight. It is the interim mitigation Horizon3 recommends and needs no application change.

  3. Take the EBS web tier off the open internet. Put supplier and remote access behind a VPN or an identity aware proxy, and restrict the rest to trusted networks.

  4. Hunt back to 20 Oct. Search web logs for external POSTs to the TCF servlet, and check for new files in the temporary directory and shells started by the application account.

  5. Block outbound internet traffic from EBS servers that do not need it. An implant that cannot call home is far less useful to an attacker.