The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
Security researchers from SafeBreach discovered a vulnerability in Google Gemini's Android voice assistant, where crafted notifications from apps like WhatsApp, Slack, and SMS could inject malicious prompts. This allowed attackers to manipulate Gemini, potentially leading to unauthorized actions, privacy breaches, and data exfiltration. Google has since patched the flaw.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves an AI system (Google Gemini's voice assistant) whose malfunction or exploitation (prompt injection vulnerability) could directly lead to significant harms including unauthorized physical control of devices, privacy violations, and impersonation attacks. These harms fall under injury or harm to persons (privacy and security breaches) and harm to property or communities (unauthorized control of smart home devices). Since the vulnerability was exploited in demonstrations and poses a direct risk of harm, this qualifies as an AI Incident. The patching and responsible disclosure are complementary information but do not negate the incident classification.[AI generated]