Skip to navigation

Threat reportVulnerabilityTL-2026-0823

FortiSandbox Unauthenticated RCE Chain: JRPC API Path-Traversal Auth Bypass and OS Command Injection (CVE-2026-39813, CVE-2026-39808, CVE-2026-25089)

criticalACTIVE

Three critical FortiSandbox flaws enable unauthenticated remote code execution on a malware-analysis security appliance: a JRPC API path-traversal authentication bypass (CVE-2026-39813), an OS command injection on the tracer-behavior API endpoint (CVE-2026-39808), and a second-order OS command injection in the web UI 'start vnc' feature affecting FortiSandbox, FortiSandbox Cloud, and FortiSandbox PaaS (CVE-2026-25089).

Key facts

IdentifiersTL-2026-0823, CVE-2026-39813, CVE-2026-39808, CVE-2026-25089
SeverityCRITICAL
CVSS9.8 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H)
Exploitation statusLinked CVE exploited in the wild; Listed in CISA KEV: CVE-2026-39808 (added ), CVE-2026-25089 (added )
EPSSCVE-2026-25089: 76.1% (99.5th percentile); CVE-2026-39808: 47.3% (98.8th percentile); CVE-2026-39813: 0.7% (52.5th percentile)
Affected productsFortinet FortiSandbox 4.4; Fortinet FortiSandbox 5.0; Fortinet FortiSandbox 4.2; Fortinet FortiSandbox Cloud 5.0; Fortinet FortiSandbox PaaS 5.0
Weaknesses (CWE)CWE-78, CWE-22, CWE-24
ToolsAI-generated ('vibecoded') CVE-2026-25089 exploit
MITRE ATT&CK techniquesT1005 Data from Local System, T1007 System Service Discovery, T1059 Command and Scripting Interpreter, T1068 Exploitation for Privilege Escalation, T1082 System Information Discovery, T1083 File and Directory Discovery, T1190 Exploit Public-Facing Application, T1211 Exploitation for Stealth, T1212 Exploitation for Credential Access, T1505 Server Software Component, T1567 Exfiltration Over Web Service, T1587 Develop Capabilities and 2 more in the attack flow
MotivationUNKNOWN
Target sectorsfinancial, government, technology, managed-security-services
Target regionsMiddle East, Global
Detection rules9 (1 shown free)
Sources16 cited sources

Detection coverage

As of 2026-06-16, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-0823 across Splunk SPL, Microsoft KQL and Sigma, covering 15 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: FortiSandbox tracer-behavior 'jid' OS Command Injection — Execution (CVE-2026-39808)

Detects CVE-2026-39808 exploitation: a GET to /fortisandbox/job-detail/tracer-behavior whose 'jid' parameter carries shell metacharacters (| ; ` $ &) or the public-PoC pipe-breakout pattern '|(' with output redirected into the web root (e.g. jid=|(id > /web/ng/out.txt)|). The jid value is passed unsanitized into a backend OS command executed as root, so any shell metacharacter or /web/ng/ reference in the jid query string is a command-injection indicator.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1190, T1059, T1059.004
  • Data sources: fortisandbox, fortisandbox:webserver, web_proxy, ids
  • Severity: critical
  • Confidence: high

Splunk SPL

index=fortisandbox sourcetype="fortisandbox:webserver" uri_path="/fortisandbox/job-detail/tracer-behavior"
| eval q=urldecode(coalesce(uri_query, cs_uri_query))
| where match(q, "jid=.*[|;`$&]") OR like(q, "%jid=|(%") OR like(q, "%/web/ng/%")
| stats count min(_time) as first_seen max(_time) as last_seen values(q) as injected_queries by src_ip dest_ip http_status
| `ctime(first_seen)` | `ctime(last_seen)`

Microsoft KQL

FortiSandbox_CL
| where TimeGenerated > ago(30d)
| where UriPath_s == "/fortisandbox/job-detail/tracer-behavior"
| extend q = url_decode(UriQuery_s)
| where q matches regex @'jid=.*[|;`$&]' or q has "jid=|(" or q has "/web/ng/"
| summarize Count = count(), FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated), Queries = make_set(q) by SrcIp_s, DstIp_s, HttpStatus_d

Sigma

title: FortiSandbox tracer-behavior jid OS Command Injection (CVE-2026-39808)
id: b2e4d3c5-0823-4c22-8f12-fb39808spl03
status: experimental
description: Detects OS command injection against the FortiSandbox tracer-behavior endpoint via shell metacharacters in the 'jid' parameter, including the public PoC pipe-breakout staging output to the web root (CVE-2026-39808).
references:
    - https://fortiguard.fortinet.com/psirt/FG-IR-26-100
    - https://github.com/samu-delucas/CVE-2026-39808
author: Threadlinqs Team
date: 2026-06-16
tags:
    - attack.initial_access
    - attack.t1190
    - attack.execution
    - attack.t1059
    - attack.t1059.004
    - cve.2026.39808
logsource:
    category: webserver
    product: fortisandbox
detection:
    selection:
        cs-uri-stem|contains: '/fortisandbox/job-detail/tracer-behavior'
        cs-uri-query|contains:
            - 'jid=|'
            - 'jid=%7C'
            - '|('
            - '/web/ng/'
    condition: selection
fields:
    - c-ip
    - cs-uri-query
falsepositives:
    - Legitimate jid values are numeric job identifiers; shell metacharacters should never appear
level: critical

Known false positives:

  • Authorized vulnerability scanners or red-team replays of the CVE-2026-39808 PoC
  • Rare URL-encoding artifacts — verify the decoded jid value before escalating

Tuning notes: Legitimate jid values are numeric job identifiers, so any non-numeric/metacharacter content is high-confidence malicious. URL-decode before matching to catch %7C/%24%28 encodings. Pair with TL-2026-0823-KQL-002 (web-root file creation) to confirm successful execution.

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)
  • FortiSandbox Web/Sandbox Service Spawns Shell — Command Injection Execution (CVE-2026-39808/CVE-2026-25089) · unlock with TL Creds
  • FortiSandbox Web-Root Output Staging & Retrieval — Web Shell / Exfil (CVE-2026-39808) · unlock with TL Creds
  • FortiSandbox 'start vnc' Second-Order Command Injection — C2/Lateral Enablement (CVE-2026-25089) · unlock with TL Creds
  • FortiSandbox Full Unauthenticated RCE Chain — Auth Bypass to Command Injection (CVE-2026-39813 + CVE-2026-39808) · unlock with TL Creds
  • FortiSandbox Appliance Defense Evasion — Verdict Tampering & Log Manipulation (Post-Compromise) · unlock with TL Creds
  • FortiSandbox Multi-CVE Probing Burst & Outbound C2 — Lateral/Exfil Enablement (CVE-2026-39813/39808/25089) · unlock with TL Creds
  • FortiSandbox JRPC API Path-Traversal Authentication Bypass — Exploitation (CVE-2026-39813) · unlock with TL Creds
  • FortiSandbox Post-Bypass JRPC Sensitive Read Methods — Discovery & Collection (CVE-2026-39813) · unlock with TL Creds

Indicators of compromise

As of 2026-06-16, Threadlinqs Intelligence tracks 15 indicators of compromise for TL-2026-0823.

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

Technical analysis

Fortinet patched the first two on 14 Apr 2026 (FG-IR-26-100, FG-IR-26-112) and the third on 9 Jun 2026 (FG-IR-26-141); Defused observed in-the-wild exploitation attempts across all three within a 24-hour window on 15-16 Jun 2026, including an AI-generated ('vibecoded') and likely faulty exploit for CVE-2026-25089.

FortiSandbox is Fortinet's network sandbox / detonation appliance used by SOCs to detonate suspicious files and URLs and render malware verdicts. Three vulnerabilities disclosed in 2026 combine to give a remote, unauthenticated attacker root-level code execution on the appliance itself — turning a defensive control into an attacker foothold and giving the adversary the ability to read, suppress, or poison malware verdicts.

CVE-2026-39813 (FG-IR-26-112, CWE-24/CWE-22 path traversal, CVSS 9.1 per Fortinet / 9.8 per NVD) is an authentication bypass in the JRPC API. The /jsonrpc/ endpoint is whitelisted in the Django middleware, so it skips the Layer-1 web authentication; Layer-2 session validation in is_valid_session() passes a user-controlled 'session' value straight into os.path.join(DIRRPCSESS, session_id) with no sanitization. Because Python's os.path.join() preserves leading '..' components, a payload of "session": "../../tmp/" resolves /usr/rpcsess/../../tmp/ to /tmp/ — a directory that always exists and whose modification time is continuously refreshed by system processes, satisfying the only two checks the validator performs (path exists, mtime within 3600s). The attacker thereby impersonates a privileged JRPC session without credentials and can invoke read methods such as sys/status (system information, 26 fields), sys/system_resource, config/scan/options, and backup/config (a ~32KB encrypted system backup).

CVE-2026-39808 (FG-IR-26-100, CWE-78 OS command injection, CVSS 9.1 per Fortinet / 9.8 per NVD) lives in an API endpoint that improperly neutralizes special elements. A public PoC targets the GET endpoint /fortisandbox/job-detail/tracer-behavior, injecting shell commands through the unsanitized 'jid' parameter using pipe characters — e.g. jid=|(id > /web/ng/out.txt)| — with output redirected into the web root (/web/ng/) for later HTTP retrieval. Commands execute with the privileges of the sandbox service (reported as root). Chained with CVE-2026-39813, an attacker can pivot from authentication bypass to full unauthenticated RCE.

CVE-2026-25089 (FG-IR-26-141, CWE-78, CVSS 9.1) is a second-order OS command injection via JSON input on the 'start vnc' feature in the web UI, allowing an unauthenticated attacker to execute unauthorized commands via crafted HTTP requests. It affects FortiSandbox 4.2 (all), 4.4.0-4.4.8, 5.0.0-5.0.5, and FortiSandbox Cloud/PaaS 5.0.4-5.0.5; it was disclosed on 9 Jun 2026, roughly a week before the active-exploitation reporting, and was credited to Adham El Karn of the Fortinet Product Security team.

Fortinet released fixes in FortiSandbox 4.4.9 and 5.0.6. As of mid-June 2026 the CVEs were not yet listed in the CISA KEV catalog and Fortinet had not formally confirmed in-the-wild compromise, but Defused/Defused-Cyber honeypot telemetry reported active exploitation attempts of all three within a single 24-hour window, with at least one CVE-2026-25089 exploit showing signs of AI ('vibecoded') authorship and being likely faulty. FortiSandbox vulnerabilities have not historically been a common attacker target, so the surge represents a notable shift. Because the appliance is a security control, post-patch hunting must include reviewing JRPC API access logs and the web UI for /jsonrpc/ requests containing '../', unusual command patterns on the tracer-behavior endpoint, suspicious files in /web/ng/, and re-validation of recent sandbox file verdicts.

MITRE ATT&CK techniques by tactic:

Collection

T1005 Data from Local System

Discovery

T1007 System Service Discovery; T1082 System Information Discovery; T1083 File and Directory Discovery

Execution

T1059 Command and Scripting Interpreter

Privilege Escalation

T1068 Exploitation for Privilege Escalation

Initial Access

T1190 Exploit Public-Facing Application

Defense Evasion

T1211 Exploitation for Stealth

Credential Access

T1212 Exploitation for Credential Access

Persistence

T1505 Server Software Component

Exfiltration

T1567 Exfiltration Over Web Service

Resource Development

T1587 Develop Capabilities

Reconnaissance

T1595 Active Scanning

defense-impairment

T1685 Disable or Modify Tools

Remediation

Patches

  • FortiSandbox 4.4.9 (FG-IR-26-100, FG-IR-26-112, FG-IR-26-141)
  • FortiSandbox 5.0.6 (FG-IR-26-112, FG-IR-26-141)

Immediate actions

  • Upgrade FortiSandbox to 4.4.9+ or 5.0.6+; migrate affected FortiSandbox Cloud/PaaS 5.0.4-5.0.5 instances to fixed 5.0.6 releases
  • Restrict network access to the FortiSandbox management interface, JRPC API (/jsonrpc/), and web UI to trusted management networks only; never expose to the internet
  • Hunt JRPC API access logs for /jsonrpc/ requests whose JSON body contains '../' or a 'session' value with path-traversal sequences
  • Search the web root /web/ng/ for unexpected files (e.g. out.txt) indicating CVE-2026-39808 command-injection output staging

Workarounds

  • Disable or firewall the JRPC API and web UI from untrusted networks until patching is complete
  • Block external HTTP/HTTPS access to /fortisandbox/job-detail/tracer-behavior and /jsonrpc/ at the perimeter

Longer-term hardening

  • Place security-management appliances behind a jump host / PAM with MFA and segment them from general user networks
  • Re-validate sandbox file verdicts produced before patching, as a compromised appliance could suppress or poison malware detection
  • Deploy WAF / IPS signatures for pipe-character injection on the tracer-behavior endpoint and path traversal on /jsonrpc/
  • Add FortiSandbox to the asset inventory and continuous external-attack-surface monitoring (e.g. runZero query os:="Fortinet FortiSandbox%")

Affected and fixed versions

  • Fortinet — FortiSandbox 4.4
    Vulnerable versions: 4.4.0; 4.4.1; 4.4.2; 4.4.3; 4.4.4; 4.4.5; 4.4.6; 4.4.7; 4.4.8
    Fixed in: 4.4.9
  • Fortinet — FortiSandbox 5.0
    Vulnerable versions: 5.0.0; 5.0.1; 5.0.2; 5.0.3; 5.0.4; 5.0.5
    Fixed in: 5.0.6
  • Fortinet — FortiSandbox 4.2
    Vulnerable versions: 4.2 (all versions, CVE-2026-25089)
    Fixed in: Migrate to a fixed 4.4 or 5.0 release
  • Fortinet — FortiSandbox Cloud 5.0
    Vulnerable versions: 5.0.4; 5.0.5
    Fixed in: 5.0.6
  • Fortinet — FortiSandbox PaaS 5.0
    Vulnerable versions: 5.0.4; 5.0.5
    Fixed in: 5.0.6

Timeline

  • Defenders are advised to review FortiSandbox JRPC API access logs from this date onward as the earliest suspected probing window (per Fyntralink incident guidance).
  • Fortinet publishes FG-IR-26-100 (CVE-2026-39808 OS command injection) and FG-IR-26-112 (CVE-2026-39813 JRPC API path-traversal auth bypass), releasing fixes in FortiSandbox 4.4.9 and 5.0.6.
  • Help Net Security reports the April FortiSandbox disclosures (CVE-2026-39813, CVE-2026-39808), both rated critical.
  • Greenbone publishes analysis and authenticated detection coverage for the April FortiSandbox RCE vulnerabilities (CVE-2026-39813, CVE-2026-39808) via its enterprise vulnerability tests.
  • NVD last-modifies the CVE-2026-39808 record (CWE-78, CVSS 9.8), referencing FG-IR-26-100 and a public GitHub PoC.
  • Fortinet publishes FG-IR-26-141 (CVE-2026-25089), a second-order OS command injection via the web UI 'start vnc' feature affecting FortiSandbox, Cloud, and PaaS; credited to Adham El Karn of Fortinet Product Security.
  • Defused honeypot telemetry observes in-the-wild exploitation attempts across all three FortiSandbox CVEs within a 24-hour window, including an AI-authored ('vibecoded'), likely faulty exploit for CVE-2026-25089.
  • runZero and Sangfor FarSight Labs publish asset-discovery and technical deep-dive guidance to help defenders locate exposed FortiSandbox appliances (e.g. runZero query os:="Fortinet FortiSandbox%") and dissect the CVE-2026-39813 path-traversal / is_valid_session() bypass mechanics.
  • Help Net Security, BleepingComputer, and The Hacker News report active exploitation attempts; a public PoC for CVE-2026-39808 is available on GitHub. Fortinet has not formally confirmed compromise and the CVEs are not yet in the CISA KEV catalog.

Sources cited

Showing 15 of 16 sources, primary sources first.

Other reports on CVE-2026-39813

Previous and next in vulnerability

Further reading

Drafted by the Threadlinqs research pipeline from 16 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