NVIDIA NeMoClaw contains an improper access control flaw in the sandbox environment initialization component. A remote attacker can send prompt‑injected content that causes the agent to read and exfiltrate host environment variables that should have been restricted during sandbox creation. The primary consequence is the disclosure of sensitive information gleaned from those variables, potentially revealing configuration secrets or credentials. This weakness is classified as CWE‑497 and results in a high CVSS score of 8.6.

The vulnerability affects NVIDIA NeMoClaw. No specific product versions are listed in the CNA data, so any deployed instance of NeMoClaw could be impacted until an official fix is applied.

The CVSS score of 8.6 indicates a high severity vulnerability requiring prompt attention. While the EPSS score is not available, the lack of a KEV listing does not mitigate the risk of exploitation. The likely attack vector is remote; an adversary must be able to interact with the NeMoClaw agent, typically via network or API access, to inject the malicious prompt that triggers the unauthorized read of host environment variables. Successful exploitation leads to unprivileged information disclosure that could aid further attacks.