Skip to navigation

Threat reportVulnerabilityTL-2026-6393

CVE-2026-3854: GitHub Enterprise Server RCE via Git Push Option Metadata Injection

highMONITORING

Semicolons in git push-option values were not neutralized before being copied into the semicolon-delimited internal X-Stat header, letting any authenticated user with push access override server-set fields (rails_env, custom_hooks_dir, repo_pre_receive_hooks) and run commands as the git service user. Wiz Research reported the same flaw on GitHub.com, which GitHub fixed on 2026-03-04; GitHub reports no exploitation beyond the researchers' own testing. First disclosed .

Key facts

IdentifiersTL-2026-6393, CVE-2026-3854
Also known asGHSA-64fw-jx9p-5j24, GitHub git push RCE, X-Stat header injection
SeverityHIGH
CVSS8.7 (CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N)
First disclosed
EPSSCVE-2026-3854: 1.5% (73.5th percentile)
Affected productsGitHub GitHub Enterprise Server; GitHub GitHub.com, Enterprise Cloud, Enterprise Cloud with Data Residency, Enterprise Cloud with Enterprise Managed Users
Weaknesses (CWE)CWE-77
MITRE ATT&CK techniquesT1005 Data from Local System, T1059 Command and Scripting Interpreter, T1078 Valid Accounts, T1190 Exploit Public-Facing Application, T1213.003 Code Repositories, T1552.001 Credentials In Files
MotivationUNKNOWN
Target sectorstechnology, software-development, enterprise
Target regionsGlobal
Detection rules9 (1 shown free)
Sources8 cited sources

Detection coverage

As of 2026-11-19, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-6393 across Splunk SPL, Microsoft KQL and Sigma, covering 3 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: CVE-2026-3854 GHES Push-Option X-Stat Field Injection (Semicolon + Injected Key) - Exploitation

Detects git push events in the GitHub Enterprise Server audit log (/var/log/github-audit.log) where a push option contains a semicolon immediately followed by an internal X-Stat key such as rails_env, custom_hooks_dir or repo_pre_receive_hooks. This is the delimiter-injection primitive of CVE-2026-3854 (CWE-77) that lets an authenticated user with push access override server-set fields and reach unsandboxed pre-receive hook execution. Based on the chain described by Wiz Research; no public PoC or in-the-wild exploitation has been reported, so treat any hit as high priority.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1190, T1078
  • Data sources: github_enterprise_audit_log (/var/log/github-audit.log), splunk index=github sourcetype=github:enterprise:audit, Syslog (GHES audit log forwarding)
  • Severity: high
  • Confidence: high

Splunk SPL

index=github sourcetype="github:enterprise:audit" earliest=-24h action="git.push"
| rex field=_raw max_match=0 "(?i);[ ]*(?<xstat_key>rails_env|custom_hooks_dir|repo_pre_receive_hooks|large_blob_rejection_enabled|reject_sha_like_refs|user_operator_mode|push_option_count)[ ]*="
| where isnotnull(xstat_key)
| eval path_traversal=if(like(_raw,"%../%"),"yes","no")
| stats count AS pushes earliest(_time) AS first_seen latest(_time) AS last_seen values(xstat_key) AS injected_xstat_keys values(path_traversal) AS path_traversal values(repo) AS repos BY host, actor, actor_ip
| convert ctime(first_seen) ctime(last_seen)
| sort - pushes

Microsoft KQL

Syslog
| where TimeGenerated > ago(24h)
| where SyslogMessage has "git.push"
| where SyslogMessage matches regex @"(?i);[ ]*(rails_env|custom_hooks_dir|repo_pre_receive_hooks|large_blob_rejection_enabled|reject_sha_like_refs|user_operator_mode|push_option_count)[ ]*="
| extend Actor = extract(@'"actor":"([^"]+)"', 1, SyslogMessage), ActorIp = extract(@'"actor_ip":"([^"]+)"', 1, SyslogMessage), Repo = extract(@'"repo":"([^"]+)"', 1, SyslogMessage), InjectedKeys = tostring(extract_all(@"(?i);[ ]*(rails_env|custom_hooks_dir|repo_pre_receive_hooks|large_blob_rejection_enabled|reject_sha_like_refs|user_operator_mode|push_option_count)[ ]*=", SyslogMessage))
| summarize Pushes = count(), FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated), Repos = make_set(Repo, 20), InjectedKeySets = make_set(InjectedKeys, 10) by Computer, Actor, ActorIp
| order by Pushes desc

Sigma

title: GHES Git Push Option Injects X-Stat Fields (CVE-2026-3854)
id: 6f1c2a7e-3b84-4d59-9a10-5e2b7c9d4a01
status: experimental
description: Detects git push audit events whose push options contain a semicolon followed by a known internal X-Stat key (rails_env, custom_hooks_dir, repo_pre_receive_hooks and others), the delimiter-injection primitive used for CVE-2026-3854.
references:
    - https://www.wiz.io/blog/github-rce-vulnerability-cve-2026-3854
    - https://github.blog/security/securing-the-git-push-pipeline-responding-to-a-critical-remote-code-execution-vulnerability/
author: Threadlinqs Team
date: 2026-11-19
tags:
    - attack.initial-access
    - attack.t1190
    - attack.t1078
    - cve.2026-3854
logsource:
    product: github
    service: audit
detection:
    selection:
        action: 'git.push'
    keywords:
        - ';rails_env='
        - ';custom_hooks_dir='
        - ';repo_pre_receive_hooks='
        - ';large_blob_rejection_enabled='
        - ';reject_sha_like_refs='
        - ';user_operator_mode='
        - ';push_option_count='
    condition: selection and keywords
falsepositives:
    - Security testing of the patched or unpatched appliance by an authorised red team
    - Push options that legitimately embed a semicolon followed by text resembling an X-Stat key name
level: high

Known false positives:

  • Authorised red-team or vendor testing of CVE-2026-3854 against a GHES appliance
  • A push option whose free-text value legitimately contains a semicolon followed by a string resembling an X-Stat key name

Tuning notes: Verify that GHES audit log forwarding includes git.push events with push options and adjust the index, sourcetype and field names (actor, actor_ip, repo) to your onboarding. Allow-list known red-team accounts and source addresses. Correlate hits with the pre-receive and gitrpcd process detections in this set. After patching to a fixed build (3.14.25, 3.15.20, 3.16.16, 3.17.13, 3.18.7, 3.19.4, 3.20.0) keep the rule enabled to catch scanning attempts.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • CVE-2026-3854 GHES Any Semicolon in Git Push Options (Broad Hunt) - Exploitation · unlock with TL Creds
  • CVE-2026-3854 GHES Shell or Downloader Spawned by gitrpcd / pre-receive / babeld - Post-Exploitation · unlock with TL Creds
  • CVE-2026-3854 GHES Bulk Repository Clone/Fetch by One Actor - Cross-Repository Collection · unlock with TL Creds
  • CVE-2026-3854 GHES Path Traversal in Injected Hook Path (repo_pre_receive_hooks / custom_hooks_dir) - Exploitation · unlock with TL Creds
  • CVE-2026-3854 GHES Git Account Reading Secret or Config Files with Interactive Tools - Post-Exploitation · unlock with TL Creds
  • CVE-2026-3854 GHES Git Account Download Tooling or Reverse Shell Patterns - C2 · unlock with TL Creds
  • CVE-2026-3854 GHES Git Service Account Running Multiple Interpreter and Recon Tools - Post-Exploitation · unlock with TL Creds
  • CVE-2026-3854 GHES Git Service Account Outbound Connection to Public IP - C2 / Lateral Movement · unlock with TL Creds

Indicators of compromise

As of 2026-11-19, Threadlinqs Intelligence tracks 3 indicators of compromise for TL-2026-6393.

3 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

CVE-2026-3854 is an improper-neutralization (CWE-77) flaw in the GitHub git push pipeline. A push passes through babeld (git proxy and entry point, which builds the internal X-Stat header), gitauth (authentication, push-access checks and session security policies) and gitrpcd (internal RPC server that performs no authentication of its own and trusts every X-Stat field). X-Stat is a semicolon-delimited key=value header; parsers split on ';' and store results in a map with last-write-wins behavior. babeld copied user-supplied push option values (encoded as push_option_0, push_option_1 ... plus push_option_count) into the header without sanitizing semicolons, so a crafted value could add new fields that override server-set ones.

Per Wiz Research, the exploit chain uses three injections: (1) a non-production rails_env value that moves pre-receive hook execution from the sandboxed path to an unsandboxed path; (2) custom_hooks_dir to control the base directory for hook lookup; and (3) a repo_pre_receive_hooks entry whose script path uses traversal to resolve to an arbitrary binary already on the server, which then runs as the git user. Other injectable fields named by Wiz include large_blob_rejection_enabled, reject_sha_like_refs, user_operator_mode and an enterprise-mode boolean. That boolean defaults to true on GHES but false on GitHub.com, so on GitHub.com it had to be injected as an additional flag to reach the custom hooks path. The pre-receive hook is a compiled Go binary that enforces policy and runs admin-defined custom hooks. Seqrite characterizes this as a trust-boundary and data-format failure rather than classic shell injection. Exploitation needs only push access: no admin rights, malicious commit content or victim interaction.

Impact on GHES is code execution as the git service user (not root), which Wiz says gave full control of the instance, including all hosted repositories and internal secrets. On GitHub.com the flaw gave code execution on shared storage nodes; Wiz enumerated millions of repository index entries on two nodes belonging to other users and organizations, but states it did not access other tenants' repository contents and validated using its own test accounts. GitHub's investigation found the exploit forces a code path normal github.com operations never use; all logged occurrences traced to Wiz's testing, and no customer data was accessed, modified or exfiltrated. GitHub fixed github.com on 2026-03-04 (root cause identified 17:45, fix deployed 19:00 per GitHub) and shipped GHES patches, then disclosed publicly on 2026-04-28. Wiz reported about 88% of GHES instances were still vulnerable at disclosure. NVD scores the flaw CVSS v3.1 8.8 (AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H); the GitHub CNA scores it CVSS v4.0 8.7. GitHub also removed an unneeded code path, left over from a different product configuration, from its container images as defense in depth. Reporting on exploitation: no in-the-wild exploitation and no public PoC have been reported in the sources reviewed; Wiz and Seqrite describe the chain without a working exploit string. Source discrepancy: Wiz and the Spanish NVD text list older fixed builds (3.14.24, 3.15.19, 3.16.15, 3.17.12, 3.18.6, 3.19.3), and The Hacker News lists 3.18.8; the GitHub post, English NVD data and Seqrite agree on the fixed versions used here. Defender guidance (GitHub, Seqrite): upgrade, review /var/log/github-audit.log for push operations whose push options contain a semicolon, correlate by account, repository, source address and time, look for unusual process execution or file access, preserve logs before changes, and rotate potentially exposed credentials and internal secrets.

MITRE ATT&CK techniques by tactic:

Collection

T1005 Data from Local System; T1213.003 Code Repositories

Execution

T1059 Command and Scripting Interpreter

Initial Access

T1078 Valid Accounts; T1190 Exploit Public-Facing Application

Credential Access

T1552.001 Credentials In Files

Remediation

Patches

  • GHES 3.14.25
  • GHES 3.15.20
  • GHES 3.16.16
  • GHES 3.17.13
  • GHES 3.18.7
  • GHES 3.19.4
  • GHES 3.20.0 or later

Immediate actions

  • Upgrade GitHub Enterprise Server to a fixed build: 3.14.25, 3.15.20, 3.16.16, 3.17.13, 3.18.7, 3.19.4 or 3.20.0 and later
  • Review /var/log/github-audit.log for push operations whose push options contain a semicolon and correlate by account, repository, source address and time
  • Preserve logs and system state before making changes if suspicious pushes are found

Longer-term hardening

  • Rotate potentially exposed credentials and internal secrets if suspicious activity is found
  • Audit how user-controlled input flows through internal protocols, especially where security-critical configuration is derived from shared data formats
  • Restrict push access to repositories to the minimum set of accounts needed

Affected and fixed versions

  • GitHub — GitHub Enterprise Server
    Vulnerable versions: 3.14.0 through 3.14.24; 3.15.0 through 3.15.19; 3.16.0 through 3.16.15; 3.17.0 through 3.17.12; 3.18.0 through 3.18.6; 3.19.0 through 3.19.3
    Fixed in: 3.14.25; 3.15.20; 3.16.16; 3.17.13; 3.18.7; 3.19.4; 3.20.0
  • GitHub — GitHub.com, Enterprise Cloud, Enterprise Cloud with Data Residency, Enterprise Cloud with Enterprise Managed Users
    Vulnerable versions: Prior to 2026-03-04
    Fixed in: Patched server-side on 2026-03-04; no customer action required

Timeline

  • GitHub deploys the fix to github.com and GitHub Enterprise Cloud variants at about 7:00 p.m.; no customer action required.
  • GitHub identifies the root cause at 5:45 p.m. and begins a forensic investigation.
  • Wiz Research discovers the flaw, confirms RCE on GHES 3.19.1 and reports it to GitHub through its Bug Bounty program; GitHub reproduces it within about 40 minutes.
  • CVE-2026-3854 is published in NVD (GitHub CNA CVSS v4.0 8.7; NVD CVSS v3.1 8.8) and GHES patches are released.
  • Wiz and GitHub publish coordinated disclosure with advisory GHSA-64fw-jx9p-5j24; Wiz reports about 88% of GHES instances still vulnerable.
  • GitHub updates its blog post; its investigation reports no exploitation beyond Wiz's own testing and no customer data accessed.
  • Seqrite publishes an analysis framing the flaw as an internal metadata injection primitive and recommending a review of the audit log for semicolons in push options.

Sources cited

Other reports on CVE-2026-3854

Drafted by the Threadlinqs research pipeline from 8 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats