CTIPilot
‹Sat · 12 Sep 2026
All daily briefs ↗
Daily brief · UTC day

Saturday, 12 September 2026

4 verified findings from 1 run · 2 updates to prior coverage · the settled record for this UTC day, in the classic brief order.

ACT NOW · CRITICALCVE-2026-42016 +1 · exploited · 7 sources · 12 Sep 04:09Z

Two dormant JFrog Artifactory bugs, patched weeks ago, are now confirmed chained into full admin takeover

Wiz Research documents in-the-wild exploitation, running 15 August–8 September 2026, of two distinct JFrog Artifactory flaws, CVE-2026-42018 (an unauthenticated caller can obtain an internal anonymous-user token) and CVE-2026-42016 (that token's scope is not validated when exchanged for a new one), chained into an unauthenticated-to-admin escalation in two HTTP requests, in some cases reaching a created administrator account in under five minutes. Both CVEs were disclosed and patched by JFrog in July and August, weeks before this exploitation was reported; six weeks after disclosure, 59% of organizations running Artifactory remained vulnerable to CVE-2026-42016, and CVE-2026-42018 fell only from 69% to 62% over four weeks.

Wiz Research confirms active, ongoing exploitation of this two-CVE chain against self-hosted JFrog Artifactory since 15 August 2026, reaching full administrative control and, in multiple cases, a dropped Rust-based backdoor with C2 capability, and six weeks after disclosure, most organizations running Artifactory still have at least one vulnerable instance. Patch every branch to its own fixed release now, 7.133.11 for CVE-2026-42016, and the CVE-2026-42018 fix matching your branch (7.111.20 / 7.117.27 / 7.125.19 / 7.133.28 / 7.146.8), since a build already above 7.133.11 numerically can still be unpatched for the second CVE. Then search access logs for the chain's own signature: an unauthenticated POST to /access/api/v1/aws/token/ followed within minutes by a POST to /access/api/v1/tokens from the same client. Any hit is a confirmed compromise, not a near-miss, revoke all tokens, audit for administrator accounts you did not create, and check the Groovy plugins directory for anything unauthorized.

Open the full advisory to act →
Criticality
Kind
Topic
Region
TL;DR · the day in one read
  1. 01Two dormant JFrog Artifactory bugs, patched weeks ago, are now confirmed chained into full admin takeover. Wiz Research documents in-the-wild exploitation, running 15 August–8 September 2026, of two distinct JFrog Artifactory flaws, CVE-2026-42018 (an unauthenticated caller can obtain an internal anonymous-user token) and CVE-2026-42016 (that token's scope is not validated when exchanged for a new one), chained into an unauthenticated-to-admin escalation in two HTTP requests, in some cases reaching a created administrator account in under five minutes. Both CVEs were disclosed and patched by JFrog in July and August, weeks before this exploitation was reported; six weeks after disclosure, 59% of organizations running Artifactory remained vulnerable to CVE-2026-42016, and CVE-2026-42018 fell only from 69% to 62% over four weeks. →
  2. 02GitLab's maximum-severity file-read flaw went from patch to in-the-wild probing in about 24 hours. GitLab shipped 19.3.2, 19.2.6 and 19.1.8 on 2026-09-10, fixing CVE-2026-85706 (CVSS 10.0): improper path confinement and missing authentication enforcement in the repository commits API let an unauthenticated user read arbitrary files from the GitLab server. Every self-managed CE/EE install from 18.7 up to those releases is affected; watchTowr's global honeypot network recorded behavioral exploitation probes against the flaw within roughly a day of disclosure, and CISA added it to its Known Exploited Vulnerabilities catalog on 2026-09-11. →
  3. 03ConnectWise patches a ScreenConnect flaw Huntress had already watched turn every infected connection into the next one's launch point. CVE-2026-84869 (CVSS 9.9) lets file-transfer actions inside an already-active ConnectWise ScreenConnect remote session run without the authorization or Host-confirmation step the product is meant to require. Huntress documented exploitation from 20 August 2026, over two weeks before ConnectWise's 8 September patch (26.6.5), in which a modified ScreenConnect client automatically pushes the same infection chain to every newly connected Host session, giving the bug a worm-like spread across an MSP's or IT department's own remote-support connections. →

01Active threats, incidents & disclosures1 item

NOTABLENATOB1

Japan's Digital Agency: a VPN vulnerability exploited since May went undetected for a month, surfaced only by an anomalous mass file-access alert on a maintenance account, exposing ~246,000 government-personnel records

Japan's Digital Agency confirmed on 2026-09-11 that Government Solution Service (GSS) (the shared PC, network and authentication environment it provisions to government ministries, independent administrative agencies and their contractors) suffered an intrusion that may have exposed personal data on roughly 246,000 individuals (Jiji Press, 2026-09-11): about 236,000 names, 231,000 email addresses, 94,000 phone numbers and 1,000 addresses, with duplication across fields (Piyolog, 2026-09-11). The agency states no National ID, bank-account or pension data was involved, and no general citizen data, only GSS-using-agency staff, associated public servants, and contracted businesses (Rocket Boys Security Measures Lab, 2026-09-11). No secondary misuse has been confirmed as of the disclosure date (Piyolog, 2026-09-11).

The root cause was a third party exploiting a vulnerability in an externally-facing VPN appliance used for maintenance access, gaining a foothold from around late May 2026 (Piyolog, 2026-09-11). The intrusion was not detected from the VPN compromise itself: on 25 June, the agency detected suspicious access from the account of a system maintenance administrator, and only through the subsequent investigation confirmed on 9 July that an external party had repeated unauthorized access since late May (Jiji Press, 2026-09-11). The agency disabled the account and cut the compromised device's external connectivity the same day, then patched the VPN appliance as an initial response (Piyolog, 2026-09-11); roughly two and a half months of investigation with an external forensics firm preceded the public announcement.

At the 2026-09-11 press conference, Digital Minister Matsumoto stated the exploited vulnerability was already known to the agency before the intrusion, rated only "Medium" severity under CVSS, and was being remediated on a severity-based schedule when it was exploited ahead of that fix being applied (Digital Agency Q&A, relayed by Piyolog, 2026-09-11). The agency stated it will review its vulnerability-management approach as a result, without disclosing specifics on what will change (Rocket Boys Security Measures Lab, 2026-09-11).

The agency said it detected suspicious access from the account of a system maintenance administrator on June 25. It then began investigating the incident and found on July 9 that an external third party had repeated unauthorized access since around late May.

"We take it seriously that the incident occurred despite our operations under multi-layered security measures and a 24-hour-a-day, 365-day-a-year surveillance system," Chief Cabinet Secretary Minoru Kihara said at a press conference

Jiji Press (via Nippon.com) 2026-09-11

悪用された脆弱性の共通脆弱性評価システム(CVSS)における評価は重要度「中(Medium)」程度で、修正プログラムやパッチの適用前に悪用された (translated from Japanese: the exploited vulnerability's CVSS severity rating was around "Medium", and it was exploited before a fix or patch was applied)

Piyolog (Piyokango), Japanese security incident-tracking blog 2026-09-11
incident12 Sep 04:09Zmulti-sourceOpen finding ↗
Sources: Jiji Press (via Nippon.com) · Piyolog (Piyokango), Japanese security incident-tracking blog · Rocket Boys Security Measures Lab
HIGHCVE-2026-84869exploitedNATOA1

CVE-2026-84869, ConnectWise ScreenConnect: a missing file-transfer authorization check lets an active remote session push and auto-run files on the Host, and Huntress traced worm-like exploitation back to 20 August, weeks before any patch existed (CVSS 9.9)

ConnectWise's own disclosure describes a client-side condition in ScreenConnect Support and Access sessions where file-transfer actions could be processed through an active remote session "without proper authorization or Host confirmation," letting files be transferred to and executed on the Host system, including through elevated execution paths (ConnectWise, 2026-09-08). ScreenConnect servers are not affected, only client versions before 26.6.5, the release that fixes CVE-2026-84869 (CWE-862 Missing Authorization, CWE-269 Improper Privilege Management, CVSS 9.9).

Huntress had been tracking the exploitation for two and a half weeks before any advisory existed. Starting 20 August 2026, its SOC found rogue ScreenConnect clients deployed via social engineering (a fake tech-support Quick Assist session in one case, a phishing-driven MSI in another) that immediately spawned four chained VBScript stagers via wscript.exe. The stagers profile the host (existing ScreenConnect install, installed security-product names, available RAM to rule out a sandbox), then pull an AES-encrypted payload bundle keyed to that profile from Dropbox. Depending on the profile, the final PowerShell stage installs a ScreenConnect backdoor client concealed as a hidden Windows service, its registry Uninstall entry removed and a restrictive service security descriptor applied, builds a UAC-bypass helper via a hijacked ms-settings: protocol handler routed through ComputerDefaults.exe, sets AmsiUtils.amsiInitFailed = true to blind AMSI, adds all of C:\Users as a Defender exclusion, and disables Defender reporting, notifications and Hypervisor-Protected Code Integrity, the last of these via a vulnerable WinRing0 kernel driver dropped alongside the payload (Huntress, 2026-09-03). The broadest observed variant then drops a wstunnel tunneling tool disguised as Themes.exe and an XMRig cryptominer disguised as SearchIndex.exe.

The installed backdoor client is what turns the bug into a worm: it continuously inspects ScreenConnect's own connection-status collection for newly established Host sessions and, on each new connection, packages the same four VBScript stagers into a ScreenConnect file-transfer message flagged to auto-run, the exact abuse path CVE-2026-84869 patches. A technician's ScreenConnect client that has been infected this way silently re-infects every subsequent host it connects to, independent of the original social-engineering vector (Huntress, 2026-09-03). No named threat-actor cluster is attributed by any source; CISA added the CVE to its Known Exploited Vulnerabilities catalog on 2026-09-11 with a three-day remediation deadline (CISA KEV, catalogue version 2026.09.11).

A condition in the ScreenConnect client may allow files to be transferred and executed through an active remote session without authorization or Host confirmation in certain circumstances. ScreenConnect servers are not impacted.

ConnectWise (vendor disclosure) 2026-09-08

This creates a worm-like behavior: propagating infections over new ScreenConnect connections. Connecting to an infected ScreenConnect client can cause the server-side Host system to receive and execute the same four-stage VBScript chain.

In late August, our Security Operations Center (SOC) sent out three critical incident reports for what looked like malicious ScreenConnect installation and unexpected process execution.

Huntress 2026-09-03
vulnerability12 Sep 04:09Zmulti-sourceOpen finding ↗
Sources: ConnectWise (vendor disclosure) · Huntress · SecurityWeek · CISA KEV catalog (catalogue version 2026.09.11)
HIGHCVE-2026-85706 +2exploitedupdatedNATOA1

CVE-2026-85706, GitLab CE/EE: unauthenticated path traversal in the repository commits API reads arbitrary server files, and honeypots caught exploitation attempts one day after the patch (CVSS 10.0)

GitLab's own patch-release notes name the root cause plainly: "improper path confinement and missing authentication enforcement in the repository commits API" let an unauthenticated caller read arbitrary files from the server (GitLab, 2026-09-10). The vulnerability, CVE-2026-85706 (CVSS 10.0, scope-changed, no privileges or user interaction required), affects every self-managed CE/EE release from 18.7 up to 19.1.8, 19.2.6 and 19.3.2, the versions GitLab shipped on 2026-09-10; GitLab.com was already running the patched version, and GitLab Dedicated customers need no action. Depending on file permissions and deployment layout, an exposed file can include configuration data, secrets, tokens, SSH keys and database credentials, turning a file-read primitive into a foothold on the CI/CD and source-control plane GitLab manages (watchTowr, 2026-09-11). The flaw was reported through GitLab's HackerOne bug-bounty program by researcher s3ntago.

watchTowr reproduced the vulnerability and validated exposure across client environments within roughly a day of GitLab's patch, and by 2026-09-11 its Attacker Eye honeypot network was already recording behavioral probes against the endpoint, meaning attackers had independently reverse-engineered a working exploit within roughly 24 hours of disclosure (watchTowr, 2026-09-11). CISA added the CVE to its Known Exploited Vulnerabilities catalog on 2026-09-11 with a three-day remediation deadline (CISA KEV, catalogue version 2026.09.11), and NCSC Switzerland's own advisory, published earlier that same day, still recorded exploitation status as "unknown", illustrating how fast the status moved within a single day (NCSC Switzerland, 2026-09-11). The same release also fixed CVE-2026-87719 (CVSS 9.9), an insecure GraphQL-subscription deserialization issue that lets an authenticated user with Duo Chat access obtain Advanced Search configurations and credentials (GitLab, 2026-09-10); it is not KEV-listed or confirmed exploited, but ships in the identical maintenance window.

GitLab has remediated an issue that, under certain conditions, an unauthenticated user could have read arbitrary files from the GitLab server due to improper path confinement and missing authentication enforcement in the repository commits API.

GitLab

watchTowr Intel is already seeing behavioral probes for this vulnerability against Attacker Eye, our global honeypot network, meaning attackers have already successfully reverse engineered and reproduced the vulnerability and the countdown to indiscriminate, in-the-wild exploitation is on.

Defenders should also hunt through log files for HTTP POST requests to “/api/v4/projects/{id}/repository/commits/” URIs containing “file.path” parameters to identify potential exploitation attempts.

watchTowr 2026-09-11
Updaterun 2026-09-13T0409Z-intelcvestagsbody

NCSC Switzerland's own advisory, published 2026-09-11, independently confirms CVE-2026-87719 (CVSS 9.9, GitLab EE only) alongside CVE-2026-85706 in the same posting. GitLab's own release notes describe the flaw: "GitLab has remediated an issue that, under certain conditions, could allow an authenticated user with Duo Chat access to obtain Advanced Search instance configurations and sensitive credentials using a specially crafted GraphQL subscription argument to bypass serialization and perform server object lookup" (GitLab, 2026-09-10). Affected versions are EE only, 18.3 up to the same 19.1.8/19.2.6/19.3.2 fixed releases already named in this entry; no exploitation is confirmed for this CVE. The same release also fixed CVE-2026-88765 (CVSS 8.5, GitLab EE only, sourced from GitLab's own release notes alone with no independent confirmation found): "GitLab has remediated an issue that, under certain conditions, could allow an authenticated user to achieve remote code execution by importing a specially crafted Git project export to overflow the Unicode conversion buffer used in Advanced Search indexing" (GitLab, 2026-09-10). This flaw's affected-version floor, EE 12.3, is substantially older than the other two flaws in this release, so any EE instance that has deferred upgrades for a long period carries the largest exposure window for this specific remote-code-execution path.

Audit logs for GraphQL subscription requests with abnormal argument shapes targeting Advanced Search, and for an Advanced Search reindex operation immediately following a Git project import from an external or untrusted source, are the discriminators for these two additional flaws.

vulnerability12 Sep 04:09Zmulti-sourceOpen finding ↗
Sources: GitLab (vendor security release notes) · watchTowr · CERT-FR (ANSSI) · NCSC Switzerland, Cyber Security Hub (CSH) · CISA KEV catalog (catalogue version 2026.09.11)

03Updates to prior coverage2 items

NOTABLEupdatedNATOA2

Finland's NCSC-FI publishes an operational manufacturer checklist for the EU Cyber Resilience Act's 24h/72h/14-day/1-month reporting clock, two weeks before the 11 September 2026 go-live

First published 2026-08-29 · open finding →

Updaterun 2026-09-12T0409Z-intelsourcing_notesourcesevidencebody

ENISA confirmed on 2026-09-11 that the CRA Single Reporting Platform has deployed its initial operating capability, resolving the open point the prior update recorded. A Bitkom survey, relayed the same day by heise online, finds only 29% of German firms understand what the CRA means for their organization and that manufacturers could not register or test their reporting process before the platform went live with the reporting obligation itself.

The open point above is resolved: ENISA confirmed on 2026-09-11 that "the EU Agency for Cybersecurity (ENISA) has deployed the initial operating capability of the Single Reporting Platform (SRP)" (ENISA, 2026-09-11), manufacturers and, from 11 December 2027, open-source software stewards submit once through the platform and the receiving CSIRT disseminates the notification to every other Member State CSIRT where the affected product is available, simultaneously to ENISA. A Bitkom survey of 1,003 German firms, relayed the same day by heise online, found only 29% know what the CRA means for their own organization and a further 38% have heard of it but cannot assess the impact; Bitkom specifically criticizes that manufacturers could not register on the platform or test their reporting process before today, "because the platform only went online with the start of the reporting obligation" (translated from German) (heise online, 2026-09-11). For a Swiss manufacturer or distributor placing networked products on the EU market, or a Swiss public-sector body procuring from an EU-regulated manufacturer now subject to this clock, the practical takeaway is that the 24-hour/72-hour/14-day-or-1-month reporting chain is a live obligation rather than a target date, and a first submission may hit friction precisely because pre-launch registration and testing were not possible.

Sources: NCSC-FI / Traficom (Finnish Transport and Communications Agency) · ENISA, Single Reporting Platform (SRP) FAQ · Hogan Lovells Cadwalader (legal analysis) · European Commission, Shaping Europe's Digital Future · ENISA, Single Reporting Platform (SRP) page · ENISA · heise online
HIGHCVE-2026-15409 +1exploitedupdatedNATOA2

CVE-2026-15409, SonicWall SMA1000: unauthenticated SSRF (CVSS 10.0) chained to post-auth code injection, actively exploited

First published 2026-07-14 · open finding →

Updaterun 2026-09-12T0409Z-inteltechniquessourcesevidenceactionssourcing_notebody

Hunt.io names a first confirmed victim of the wider mass-exploitation wave against this same CVE, the Borough Council of King's Lynn and West Norfolk (UK local government), and reconstructs a technique distinct from UTA0533's earlier targeted intrusion: operators ran a standalone Linux build of Impacket's secretsdump directly on compromised SonicWall appliances against internal Windows hosts, reaching full DCSync replication in 5 of 250 targeted environments. This is reported as a materially broader, opportunistic wave, not attributed to UTA0533 or INC Ransom.

Hunt.io's analysis, relayed at length by Security Affairs, links the 17 July 2026 cyberattack on the Borough Council of King's Lynn and West Norfolk (UK local government) (with moderate confidence) to a wider mass-exploitation campaign against CVE-2026-15409, materially broader than the UTA0533-attributed targeted intrusion the earlier updates above describe (Security Affairs, 2026-09-11). Hunt.io's AttackCapture system crawled an operator's own unauthenticated open directory the same day the council disclosed, recovering the complete toolchain and campaign scope: 250 target appliances, 168 with LDAP configurations exposed (534 Active Directory accounts across 160 domains), 9 environments losing SAM/LSA secrets, and 5 losing their entire domain database via full DCSync replication against 7 domain controllers, with confirmed credential theft in France, India, Italy and the US and a broader target list spanning the UK, Canada, Germany, Sweden, Poland, Hungary, South Korea and Hong Kong. OffSeq's Threat Radar, an AI-generated summary tracker, carries the same campaign statistics in its own posting on the report (OffSeq Threat Radar, 2026-09-10), a repost rather than independent verification, since it traces to the same Hunt.io findings rather than a separate observation.

The technique this wave adds: operators ran a standalone Linux build of Impacket's secretsdump directly on the compromised SonicWall appliance itself against internal Windows systems, extracting SAM and LSA secrets and, where the harvested LDAP credentials lacked sufficient privilege to run DCSync directly, falling back to pass-the-hash DCSync replication using machine-account NTLM hashes recovered from those LSA secrets (Security Affairs, quoting Hunt.io, 2026-09-11). Running the credential-theft tooling from the appliance's own operating system, rather than from a foothold on a monitored Windows or Linux host, is the operative choice: "By 16 July 2026, two days after SonicWall disclosed CVE-2026-15409, a threat actor was using modified, multithreaded tooling to identify vulnerable SMA 1000 appliances, extract and decrypt LDAP credentials, and automate follow-on credential theft against internal Active Directory environments" (Hunt.io, quoted by Security Affairs, 2026-09-11). Scripts recovered from the directory carried extensive Chinese-language comments, which Hunt.io states is insufficient on its own for attribution; no link to UTA0533 or to INC Ransom is asserted in this reporting, treat it as a separate, broader post-disclosure scanning wave rather than the same operator.

Defender takeaway (delta only): organizations typically have far less process, file and network visibility into a VPN/firewall appliance's own operating system than into EDR-monitored Windows and Linux hosts, this wave specifically exploited that gap by running its credential-theft tooling on the appliance itself. Triage: where forensic capability allows inspecting the appliance's own filesystem and process history (not just its network logs), an unexpected standalone binary or interpreter process on the appliance (distinct from its normal service processes) run against internal LDAP/AD infrastructure is the signature this wave adds to the existing detection guidance above.

Sources: SonicWall PSIRT · Volexity · Rapid7 · The Hacker News · Resecurity · SecurityWeek · Dark Reading · SonicWall · Security Affairs · OffSeq Threat Radar

04Deep dive1 item

CRITICALCVE-2026-42016 +1exploitedNATOB2

CVE-2026-42016 + CVE-2026-42018, JFrog Artifactory: chaining two previously-patched token flaws turns an unauthenticated request into full administrative control in two API calls, confirmed exploited since mid-August

JFrog disclosed the two flaws separately and weeks apart, labeling each only "High" without a numeric score; the CNA's own submitted CVSS 3.1 base scores (8.1 for CVE-2026-42016 and 7.5 for CVE-2026-42018) are recorded in the MITRE CVE record for each (MITRE CVE Record, CVE-2026-42016; MITRE CVE Record, CVE-2026-42018). CVE-2026-42016 (CWE-863 Incorrect Authorization), disclosed 27 July 2026, is a privilege-escalation bug in Artifactory's token handling: the platform validates a token's signature and issuer but never checks whether the token's scope matches what it is being used for, so a low-privileged token can be exchanged for one carrying administrative authority (JFrog, 2026-07-27). CVE-2026-42018 (CWE-287 Improper Authentication), disclosed 12–13 August 2026, is narrower but supplies the missing first step: Artifactory returns an internal anonymous-user token to a caller who never authenticated, even when anonymous access is disabled (JFrog, 2026-08-13). Neither flaw alone grants administrative control; Wiz Research states plainly that the anonymous token from CVE-2026-42018 and the scope-validation gap in CVE-2026-42016 "together... can turn an unauthenticated request into an admin-scoped token in two steps" (Wiz Research, 2026-09-10).

Kill chain, as Wiz reconstructed it from real intrusions (initial access, T1190). An unauthenticated POST /access/api/v1/aws/token/ (the trailing slash matters; the bare path without it returns 401) returns HTTP 200 with a JWT scoped to Artifactory's internal anonymous identity, exploiting CVE-2026-42018. The operator then exchanges that JWT for a new one via POST /access/api/v1/tokens, which returns HTTP 200 carrying admin authority despite the request originating from the anonymous identity, exploiting CVE-2026-42016's missing scope check (T1550.001, forging elevated authority onto an application access token). The escalated token still shows token:anonymous as its actor in logs, which is itself a detection anomaly: no legitimate administrative action originates from the anonymous identity. Wiz observed actors reach a created administrator account in under five minutes from the first request in some cases. CISA added both CVEs to its Known Exploited Vulnerabilities catalog on 2026-09-11 (CISA KEV, catalogue version 2026.09.11).

Post-exploitation observed across the chained cases. Operators create persistent administrator accounts via PUT /api/security/users/<username> or /access/api/ui/users/<username> (T1136.001); deploy malicious Groovy plugins through Artifactory's own extensible plugin framework to gain arbitrary code execution (T1505, abusing a legitimate server extension mechanism for persistence); and run ad-hoc shell commands through the plugin-execution endpoint (GET/POST /api/plugins/execute/<plugin>), used for reconnaissance and file enumeration (T1059). Across multiple cases Wiz observed a custom Rust-based backdoor with command-and-control capability dropped to disk as a further persistence layer.

Wiz's own patching-velocity data is the second half of the defender-relevant finding: at first disclosure, 67% of organizations running Artifactory had at least one instance vulnerable to CVE-2026-42016, and a similar 69% for CVE-2026-42018; six weeks after CVE-2026-42016's disclosure, 59% remained vulnerable, and CVE-2026-42018 fell only from 69% to 62% over four weeks, versus the separately-disclosed CVE-2026-82329 (a default-configuration join-key bypass, disclosed 28 August 2026), which dropped from 67% to 49% within two weeks of its own disclosure. Wiz attributes the gap to CVE-2026-82329's critical-severity label likely driving faster security-team attention, while these two high-severity CVEs went unpatched at far higher rates for far longer; precisely the CVSS-driven prioritization trap that let this chain stay open long enough to be found under active exploitation.

Wiz Research has identified active, in-the-wild exploitation of three critical and high-severity vulnerabilities affecting JFrog Artifactory: CVE-2026-42016, CVE-2026-42018, and CVE-2026-82329. Attackers are chaining these vulnerabilities to bypass authentication, escalate privileges, and gain administrative control over vulnerable Artifactory instances.

Between August 15 and September 8, 2026, we observed multiple actors chain CVE-2026-42018 and CVE-2026-42016 against self-hosted Artifactory instances. Across multiple cases we observed a custom Rust backdoor with C2 capabilities being dropped. Wiz Research is not aware of any prior public reporting of in-the-wild exploitation involving those two CVEs.

Wiz Research 2026-09-10

JFrog Artifactory (Self Hosted) versions before 7.133.11 are vulnerable to a privilege escalation attack due to a validation check of the token signature/issuer and not the token's scope.

JFrog (vendor security advisories) 2026-07-27

Builds on: 2026-09-01/jfrog-artifactory-cve-2026-82329-default-config-admin-bypass

vulnerability12 Sep 04:09Zsingle-sourceOpen finding ↗
Sources: JFrog (vendor security advisories) · Wiz Research · BSI / CERT-Bund (WID-SEC-2026-2808) · MITRE CVE Record (CNA metrics) · CISA KEV catalog (catalogue version 2026.09.11)

05Action items5 items

Verification & coverage notes1 run

2026-09-12T0409Z-intel · Sonnet 5 · window 26 h · 4 entries published

Verification & coverage notes

Coverage window: standard (gap_hours=23.98 since 2026-09-11T0410Z-intel, window_hours=26). No outage-backfill duty.

Mechanical KEV sweep (v4.8), all 4 in-window additions dispositioned. tools/kev_window_diff.py found four CISA KEV additions dated 2026-09-11 uncovered by the store: CVE-2026-42016 and CVE-2026-42018 (JFrog Artifactory), CVE-2026-84869 (ConnectWise ScreenConnect), CVE-2026-85706 (GitLab CE/EE). All four published as new entries this run. CVE-2026-42016/-42018 were confirmed via JFrog's own advisory to be distinct root causes, disclosure dates and affected-version ranges from the already-tracked CVE-2026-82329 (2026-09-01 entry) (not a same-flaw re-numbering) so they ship as a separate entry with references[] pointing at the existing one, per the single-report-covers-three-CVEs judgment call the researching sub-agent flagged; the two entries are not merged.

New entries (4):

  • 2026-09-12/cve-2026-84869-connectwise-screenconnect-worm-file-transfer (vulnerability, high), worm-like exploitation from 2026-08-20, patch 2026-09-08, KEV 2026-09-11.
  • 2026-09-12/cve-2026-85706-gitlab-unauth-path-traversal-file-read (vulnerability, high), patch-to-honeypot-probe in ~24h, KEV 2026-09-11.
  • 2026-09-12/jfrog-artifactory-cve-2026-42016-42018-token-chain-takeover (vulnerability, critical, deep dive, category supply-chain, not used in the last 30 days), confirmed wide-scale admin-takeover chain against two previously-patched, low-attention CVEs, with backdoors already dropped and the majority of installs still vulnerable six weeks post-disclosure; raised from high to critical on iteration-2 verifier calibration flag, consistent with store precedent (the 2026-09-01 CVE-2026-82329 entry, same product, same admin-takeover outcome, critical). Single-source (Wiz Research is the sole assessor of the exploitation claim; JFrog's own advisories corroborate only the underlying flaws, not the exploitation).
  • 2026-09-12/japan-digital-agency-gss-vpn-breach-maintenance-account (incident, notable), clears the PD-11 breach gate on limb (b): a materially transferable lesson (detection via anomalous maintenance-account file-access volume rather than the initial exploit; a shift from CVSS-severity-based to risk-based patch prioritization after losing the race to an exploit on a bug rated only "Medium") on a structurally close foreign analogue (a national government's own shared multi-agency IT-services platform).

Updated entries (2):

  • 2026-08-29/eu-cra-reporting-obligation-ncsc-fi-checklist, update, floats updated_at. ENISA confirmed 2026-09-11 the CRA Single Reporting Platform has deployed initial operating capability, resolving the prior fire's open point; a same-day Bitkom survey (via heise) adds a preparedness data point.
  • 2026-07-14/sonicwall-sma1000-ssrf-cve-2026-15409-actively-exploited, update, floats updated_at. Hunt.io (via Security Affairs; OffSeq's AI-generated Threat Radar reposts the same figures, not an independent source) names a first confirmed victim of a materially wider, distinct mass-exploitation wave against the same CVE (UK council; secretsdump run directly on the compromised appliance reaching DCSync in 5/250 environments), explicitly not attributed to UTA0533 or INC Ransom.

Borderline drop: Anthropic's "distillation attacks" report (hydra-cluster fraudulent-account networks used by DeepSeek/Moonshot/MiniMax to scrape Claude), substantive primary technical research (Anthropic's own telemetry) with a transferable fraudulent-account-network detection technique, but the underlying subject is an AI-industry IP/competitive dispute under an active US-China export-control policy backdrop, with no direct Swiss public-sector nexus beyond a generic account-fraud/API-abuse technique-class analogy. Relevance doubt resolved toward drop per PD-11/v4.2 calibration. Not registered to the entity registry.

Single-source / carve-out items: jfrog-artifactory-cve-2026-42016-42018-token-chain-takeover (single-source, reliability B / credibility 2; Wiz Research is the sole assessor of the active-exploitation chain; sourcing_note explains why BSI/ENISA's same-day postings are not counted as independent corroboration).

Coverage gaps: inside-it-ch (whole-host "Vercel Security Checkpoint" 429 on specific article bodies, the long-carried Insel Gruppe/ServiceNow backlog row plus two further articles this run, a 14th+ consecutive fire unable to read this class of article; RSS and general listing access stayed healthy throughout); cert-pl (/en/posts/ news listing 403'd this run; its vuln-advisory RSS feed, the source's primary recipe, was not re-tried as duplicative of S1's own essential-tier coverage of the same source, covered_anyway: true); cert-at (reachable, no in-window items, latest post dated 2026-08-26, outside the 26h window).

Coverage backlog: all 11 open rows in state/coverage_backlog.md re-checked this run (S1–S4); every one carries a 2026-09-12 dated note, no change on any, no row struck.

No contradictions found this run. No product/supplier watchlist configured (sweep is a no-op per the profile). No closed-source intake (no intel/ drops in-window).

Verification loop, 8 iterations, fail-open at the cap (never reached a confirmed CLEAN). The loop ran the full 8-iteration cap: iteration 1 NEEDS_FIXES (truth 3, editorial 6), 2 NEEDS_FIXES (3+2), 3 NEEDS_FIXES (4+1), 4 NEEDS_FIXES (2+0), 5 CLEAN, 6 (the confirmation pass) NEEDS_FIXES (3+1), restarting the chain per decision rule 2, 7 NEEDS_FIXES (1+3), 8 NEEDS_FIXES (0+1). Every iteration's findings were genuine and were remediated (see verification.iterations[] above and work/2026-09-12T0409Z-intel/verification.iter{1..8}.md); no finding was declined without a stated reason. Per the cap fail-open rule, this run publishes on iteration 8's NEEDS_FIXES verdict with verification_residual_count: 1 (one small, evidenced missing-citation finding, fixed) rather than a ninth iteration. This is an unusually long loop for what were, in aggregate, narrow and progressively smaller defects (import citation gaps, one inverted causal clause, one unsupported given name, a few CVSS/date sourcing gaps), every one caught by a genuinely independent cold read finding something the previous seven passes missed, which is the loop working as designed rather than a sign the content is unsound. The one substantive editorial judgment call (raising jfrog-artifactory-cve-2026-42016-42018-token-chain-takeover from priority: high to critical on iteration 2's flag) was independently sanity-checked by iteration 3 and not contested by any later iteration.