
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
Researchers discovered a critical vulnerability (CVE-2026-65105) in Nvidia's NemoClaw, enabling attackers to hijack and permanently poison local AI agents via a malicious website. The flaw allows unauthorized control, data theft, and persistent manipulation of AI outputs, with Windows deployments remaining unpatched and standard security tools unable to detect the attack.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves an AI system (NemoClaw deploying AI agents using LLMs) and describes a security flaw that allows attackers to manipulate the AI system's behavior maliciously. This manipulation leads to direct harm by compromising AI agents' operations, which can result in unauthorized access, data exfiltration, and suppression of security measures. These outcomes constitute harm to property, communities, or organizational security, fitting the definition of an AI Incident. The vulnerability has been exploited in proof of concept, indicating realized harm potential rather than just a theoretical risk, and remediation is complex due to persistent poisoning of the model template. Therefore, this event qualifies as an AI Incident.[AI generated]