CVE-2026-24222
Published: 28 April 2026
Summary
CVE-2026-24222 is a high-severity Exposure of Sensitive System Information to an Unauthorized Control Sphere (CWE-497) vulnerability in Nvidia Nemoclaw. Its CVSS base score is 8.6 (High).
Operationally, exploitation aligns with the MITRE ATT&CK technique Data from Local System (T1005); ranked at the 17.0th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog.
The strongest mitigations our analysis identified are NIST 800-53 AC-25 (Reference Monitor) and SC-39 (Process Isolation).
Threat & Defense at a Glance
Threat & Defense Details
Mitigating Controls (NIST 800-53 r5)AI
Implements a reference monitor in the sandbox environment to enforce access control policies, preventing the agent from reading unauthorized host environment variables during initialization.
Validates and sanitizes prompt-injected content inputs to the sandbox initialization component, blocking exploitation that tricks the agent into accessing host environment variables.
Enforces process isolation between the sandbox agent and host system, restricting access to and exfiltration of sensitive host environment variables during sandbox creation.
MITRE ATT&CK Enterprise TechniquesAI
Why these techniques?
Vulnerability directly enables unauthorized access to host environment variables from sandbox (T1005 Data from Local System) followed by exfiltration of that data (T1041 Exfiltration Over C2 Channel).
NVD Description
NVIDIA NeMoClaw contains a vulnerability in the sandbox environment initialization component, where a remote attacker could cause improper access control by sending prompt-injected content that causes the agent to read and exfiltrate host environment variables not properly restricted during sandbox…
more
creation. A successful exploit of this vulnerability might lead to information disclosure.
Deeper analysisAI
CVE-2026-24222 affects NVIDIA NeMoClaw in its sandbox environment initialization component. The vulnerability enables improper access control (CWE-497), where a remote attacker sends prompt-injected content that causes the agent to read and exfiltrate host environment variables not properly restricted during sandbox creation. Published on 2026-04-28, it has a CVSS v3.1 base score of 8.6 (AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:N/A:N), indicating high severity due to network accessibility, low attack complexity, and significant confidentiality impact.
A remote attacker requires no privileges or user interaction to exploit this vulnerability. By crafting and sending malicious prompt-injected content, the attacker can trick the agent into accessing sensitive host environment variables during sandbox setup, leading to their exfiltration and potential information disclosure.
Mitigation details are available in official advisories, including NVIDIA's security bulletin at https://nvidia.custhelp.com/app/answers/detail/a_id/5837, the NVD entry at https://nvd.nist.gov/vuln/detail/CVE-2026-24222, and the CVE record at https://www.cve.org/CVERecord?id=CVE-2026-24222. Security practitioners should consult these for patch information and workarounds.
Details
- CWE(s)