GitLab CVE-2026-85706: new max-severity path traversal exploited one day after disclosure, mass exploitation imminent per WatchTowr, hunt POST requests to commits API with file.path parameters      JFrog three-CVE chain: Wiz confirms attackers chaining CVE-2026-42018 and CVE-2026-42016 with CVE-2026-82329 since mid-August to deploy backdoors on self-hosted Artifactory instances      CISA KEV: ConnectWise ScreenConnect CVE-2026-84869 CVSS 9.9 deadline today, MikroTik RouterOS two flaws deadline passed Saturday, JFrog two additional CVEs deadline September 25      GitLab CVE-2026-85706: new max-severity path traversal exploited one day after disclosure, mass exploitation imminent per WatchTowr, hunt POST requests to commits API with file.path parameters      JFrog three-CVE chain: Wiz confirms attackers chaining CVE-2026-42018 and CVE-2026-42016 with CVE-2026-82329 since mid-August to deploy backdoors on self-hosted Artifactory instances      CISA KEV: ConnectWise ScreenConnect CVE-2026-84869 CVSS 9.9 deadline today, MikroTik RouterOS two flaws deadline passed Saturday, JFrog two additional CVEs deadline September 25     
CyberSipTM
Intelligence without the noise
Issue No. 129
September 14, 2026
3 items · past 3 days
<5 min read
Today's picture

A new GitLab path traversal vulnerability, CVE-2026-85706, was exploited in the wild one day after its September 12 disclosure with WatchTowr warning mass exploitation is imminent, placing two critical GitLab CVEs under simultaneous active attack. Wiz confirmed that attackers have been chaining three JFrog Artifactory vulnerabilities since mid-August to deploy persistent backdoors across self-hosted instances, materially updating the CVE-2026-82329 story from Issue 122 with a multi-CVE chain and a confirmed backdoor objective. CISA added five vulnerabilities to its Known Exploited Vulnerabilities catalog across ConnectWise ScreenConnect, MikroTik RouterOS, and JFrog Artifactory, with the ScreenConnect deadline falling today and the MikroTik deadlines already passed.

Today's intelligence
3 items
01 CriticalGitLab CVE-2026-85706Mass Exploitation Imminent
A new max-severity GitLab path traversal was exploited one day after disclosure — WatchTowr says mass exploitation is imminent, with two critical GitLab CVEs now simultaneously active
CVE-2026-85706 is an unauthenticated path traversal in GitLab's repository commits API that lets any attacker read arbitrary files from the server. Disclosed September 12, exploited September 13. WatchTowr warns mass exploitation will follow. This is the second critical GitLab CVE under active attack — CVE-2026-19478 from Issue 114 remains exploited in parallel. Hunt POST requests to the commits API URI containing file.path parameters immediately.
CVECVE-2026-85706
Max severity
TypePath traversal
Unauthenticated
Arbitrary file read
Exploited1 day after
Sept 12 disclosure
Mass exploitImminent
per WatchTowr
Hunt indicatorHTTP POST to
/api/v4/projects/
{id}/repository/
commits/ URIs
with file.path
parameter
GitLab disclosed CVE-2026-85706 on September 12, a maximum-severity path traversal vulnerability in its repository commits API that allows unauthenticated attackers to read arbitrary files from the GitLab server by crafting POST requests to the /api/v4/projects/{id}/repository/commits/ endpoint containing manipulated file.path parameters. Exploitation in the wild was confirmed September 13, one day after disclosure. WatchTowr head of threat intelligence Jake Knott stated that mass exploitation is likely to follow shortly, citing the severity, the unauthenticated attack vector, and the scale of internet-exposed self-managed GitLab instances. Security firm CyCognito estimates more than 20,000 self-managed GitLab instances are reachable on standard web ports, with a meaningful share still unpatched days after critical advisories ship. This is the second simultaneous critical GitLab CVE under active attack: CVE-2026-19478, the GraphQL code injection tracked since Issue 114, remains under active exploitation in parallel. GitLab also patched CVE-2026-87719 in the same release, a CVSS 9.9 Duo Chat GraphQL serialization bypass that allows an authenticated user to obtain sensitive credentials and Advanced Search instance configurations.
GitLab is the repository and CI/CD system of record for an organization's code, pipelines, secrets, and deployment configurations. Unauthenticated arbitrary file read on a GitLab server is access to anything the server can read: source code, .env files, SSH keys, CI/CD variable files, tokens embedded in repository configuration, and GitLab's own configuration files containing database credentials and secret keys. Two critical CVEs simultaneously under active attack on the same platform doubles the attack surface and means organizations patching one but not the other remain exposed.
WatchTowr's specific hunting query is the most operationally useful indicator available before patches are confirmed applied across all instances. Hunt HTTP POST requests to /api/v4/projects/{id}/repository/commits/ URIs containing file.path parameters in GitLab web server logs. For CVE-2026-19478, the companion indicator remains requests containing @gl_introduced in the GraphQL endpoint logs. Run both queries against logs from September 12 forward given the confirmed exploitation timeline.
  • Apply the GitLab patch for CVE-2026-85706 immediately to all self-managed instances. Patch is available in the versions GitLab released September 12. For instances that cannot patch immediately, restrict public access to the repository commits API endpoint as an interim control and run the WatchTowr hunting query against existing logs.
  • Run both GitLab hunting queries now: POST requests to the commits API containing file.path parameters (CVE-2026-85706) and requests containing @gl_introduced in GraphQL endpoint logs (CVE-2026-19478). Both CVEs are under active exploitation simultaneously. Confirm patch status for both CVEs independently.
Disclosed Friday. Exploited Saturday. Mass exploitation imminent Monday. Unauthenticated file read on your code repository means source code, secrets, and CI/CD tokens. Two critical GitLab CVEs active at once. Patch both, hunt both, treat them as separate items.
02 HighJFrog ArtifactoryThree-CVE Backdoor Chain
Attackers have been chaining three JFrog Artifactory CVEs since mid-August to deploy backdoors — the CVE-2026-82329 story from Issue 122 is more severe than first understood
Wiz confirmed that CVE-2026-42018 and CVE-2026-42016, two Artifactory flaws patched July 27, have been chained with CVE-2026-82329 since mid-August. The chain: get an anonymous token via CVE-2026-42018, escalate to admin via CVE-2026-42016, then pivot to full unauthenticated admin access via CVE-2026-82329. The end objective is not credential theft but persistent backdoor installation on the Artifactory server itself. CISA added CVE-2026-42018 and CVE-2026-42016 to KEV this week.
Chain step 1CVE-2026-42018
Anon token extract
Patched July 27
Chain step 2CVE-2026-42016
Privilege escalation
to admin
Patched July 27
Chain step 3CVE-2026-82329
Auth bypass
CISA KEV
ObjectivePersistent backdoor
on Artifactory
server, not just
credential theft
Active sinceMid-August 2026
Wiz published findings confirming that attackers have been actively chaining three JFrog Artifactory vulnerabilities since mid-August 2026 to achieve persistent backdoor access to self-hosted Artifactory instances. The chain uses CVE-2026-42018, an insufficient token validation flaw patched July 27, to extract an anonymous-user token from the Artifactory instance. That token is then used with CVE-2026-42016, a privilege escalation flaw also patched July 27, to elevate the anonymous-user token to administrator-level access. With administrator access established, attackers layer in CVE-2026-82329, the authentication bypass covered in Issue 122, which provides a parallel direct route to administrator tokens. CISA added CVE-2026-42018 and CVE-2026-42016 to its Known Exploited Vulnerabilities catalog as part of the five-CVE batch this week, setting a September 25 remediation deadline. The confirmed objective is persistent backdoor deployment on the Artifactory server itself, not only credential or artifact theft. Wiz observed backdoor installations across multiple compromised self-hosted Artifactory instances between August 15 and September 8.
Issue 122 framed the CVE-2026-82329 story as attackers minting admin tokens and establishing privileged persistent access, with artifact tampering as a potential downstream risk. Wiz's finding confirms the next step actually occurred: backdoors were deployed on the server itself during a period when organizations believed patching CVE-2026-82329 was the complete remediation. Organizations that patched only CVE-2026-82329 and did not address the two July 27 flaws left half the chain available. And organizations that applied all three patches after mid-August without investigating whether a backdoor was installed before patching may be operating with undetected persistent access on their Artifactory server.
  • Verify that CVE-2026-42018 and CVE-2026-42016 are patched on all self-hosted Artifactory instances, not only CVE-2026-82329. The July 27 patches for the first two CVEs in the chain are a precondition for blocking the full escalation sequence.
  • Review Artifactory server-level indicators for backdoor artifacts installed between August 15 and September 8. Check for unexpected files, scheduled tasks, or processes associated with the Artifactory service account. A patched instance that was accessible during the mid-August to early September window should be investigated for backdoor presence before being treated as clean.
Three Artifactory CVEs chained since mid-August. The objective is a persistent backdoor on the server, not just admin tokens. If you only patched CVE-2026-82329, the other two steps in the chain remained available. If you patched all three after August 15, check for backdoor artifacts before assuming the instance is clean.
03 HighCISA KEV BatchScreenConnect + MikroTik
CISA added five CVEs to KEV this week: ScreenConnect CVSS 9.9 deadline is today, MikroTik RouterOS deadlines already passed
ConnectWise ScreenConnect CVE-2026-84869 (CVSS 9.9) allows unauthorized file transfer and code execution through any active remote session without host confirmation — federal deadline today. Two MikroTik RouterOS flaws (CVE-2026-67277 memory disclosure and CVE-2026-86060 privilege escalation) had federal deadlines of September 13. The JFrog CVE-2026-42018 and CVE-2026-42016 additions complete the five-CVE batch with a September 25 deadline.
ScreenConnectCVE-2026-84869
CVSS 9.9
KEV deadline today
MikroTik 1CVE-2026-67277
CVSS 8.8
Memory disclosure
Deadline Sept 13
MikroTik 2CVE-2026-86060
CVSS 9.2
Privilege escalation
Deadline Sept 13
JFrogCVE-2026-42018
CVE-2026-42016
Deadline Sept 25
CISA added five vulnerabilities to the Known Exploited Vulnerabilities catalog last week. CVE-2026-84869 in ConnectWise ScreenConnect is a missing authorization flaw rated CVSS 9.9 that allows an attacker to transfer files and execute them through an active remote session without requiring host confirmation or authorization from the session participant. ConnectWise is the remote access and RMM platform used extensively by MSPs and enterprise IT support teams. The federal remediation deadline is today, September 14. Two MikroTik RouterOS flaws, CVE-2026-67277 (kernel memory disclosure and remote denial-of-service in the bandwidth-test service, CVSS 8.8) and CVE-2026-86060 (privilege escalation via a command argument injection in the SSH login path that grants full administrative access with a crafted username, CVSS 9.2), had federal deadlines of September 13. Both deadlines are now passed. The JFrog additions, CVE-2026-42018 and CVE-2026-42016, are covered in story two above with a September 25 deadline.
ScreenConnect CVE-2026-84869's CVSS 9.9 and "no host confirmation" language describes an attacker who has established any legitimate or compromised ScreenConnect session and can then use that session to transfer and execute arbitrary files on the connected host without the end user being aware or able to deny the action. In MSP environments where ScreenConnect sessions to client endpoints are routine, a compromised ScreenConnect account or a hijacked session becomes direct code execution on client systems. The MikroTik RouterOS flaws affect the network routing infrastructure used extensively in enterprise branch offices and ISP networks, where privilege escalation via a crafted SSH username string (CVE-2026-86060) gives an attacker root access to the routing device with no vulnerability exploitation required beyond the login attempt itself.
  • Apply the ConnectWise ScreenConnect patch for CVE-2026-84869 today. The federal deadline is today. For MSPs using ScreenConnect to manage client environments, treat this as the same supply-chain urgency as N-central from Issue 127: a compromised ScreenConnect deployment is access to every client endpoint it connects to.
  • Update MikroTik RouterOS to address CVE-2026-67277 and CVE-2026-86060. Both federal deadlines have passed. For CVE-2026-86060 specifically, the crafted SSH username path means any RouterOS device with SSH access exposed to untrusted networks is at immediate exploitation risk and should be prioritized for patching or SSH access restriction today.
ScreenConnect deadline is today: patch it or your MSP clients are at risk. MikroTik SSH privilege escalation via a crafted username — no exploitation technique required, just a login attempt. Both MikroTik deadlines passed Saturday. Five KEV additions in one week.
Cross-source standouts
01
GitLab's second simultaneous critical CVE: what it means when two high-severity flaws in the same platform are exploited concurrently
CVE-2026-19478 (GraphQL code injection, Issue 114) and CVE-2026-85706 (commits API path traversal) are now both under active exploitation simultaneously. The operational problem this creates is patch-tracking divergence: organizations that confirmed CVE-2026-19478 was patched may not have flagged CVE-2026-85706 as a separate patch action. The two CVEs affect different GitLab components, require different patches, and have different hunting indicators. Treating GitLab as either patched or unpatched as a binary is insufficient when two separate critical CVEs are concurrently in-the-wild. Patch management teams should verify patch status for both CVEs independently and run hunting queries for both indicators, not treat one as a proxy for the other.
02
The JFrog chain and the partial-patch problem: patching the disclosed CVE while leaving the chain available
Issue 122 covered CVE-2026-82329 as the primary JFrog Artifactory story. Organizations that responded by patching CVE-2026-82329 specifically, without also addressing CVE-2026-42018 and CVE-2026-42016 from the July 27 release, left two out of three steps in the exploitation chain available. This is the same partial-patch dynamic this brief has documented for PaperCut (two emergency patches with known bypasses), ShieldBreak and ShieldCrash (patch bypass available day one), and Cisco FMC (CVE-2026-20079 patched but CVE-2026-20316 companion flaw chained in active attacks). The consistent theme is that exploitation chains are resilient to single-CVE patching when the chain spans multiple CVEs in the same platform. Verifying patch status for every CVE in a disclosed chain, not only the highest-profile one, is the operational response the JFrog situation now makes concrete.
Still watching
Days 2–7+
ShieldCrash (ShieldBreak bypass) (Issue 126 · Nightmare Eclipse, no CVE, file read as SYSTEM on latest patched Windows) — ShieldBreak CVE-2026-69414 patched. ShieldCrash unpatched. Monitor MSRC. Keep behavioral detection active for the Nightmare Eclipse series.
Day 5
SAP OVERPASS CVE-2026-44756 (Issue 126 · CVSS 10.0, unauthenticated SAP kernel EPP RCE, no exploitation confirmed) — apply patch per SAP Note 3500180. Restrict EPP endpoint to authorized networks. Window before first exploitation is short based on prior CVSS 10.0 SAP pattern.
Day 5
StyleSmuggler CVE-2026-75650 (Issue 125 · CVSS 10.0, Magento/Adobe Commerce, actively exploited since September 4) — apply composer patch VULN-39341. Rotate encryption key and all derived credentials. Run eComscan for the Linux backdoor.
Day 7+
GitLab CVE-2026-19478 (Issue 114 · GraphQL code injection, confirmed exploited, 90-day embargo for 18.2–18.10 through mid-November) — hunt @gl_introduced in GraphQL logs. Patch available for 19.x and 18.11+ only. Now compounded by CVE-2026-85706 above.
Day 7+