Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore a groundbreaking cybersecurity conference talk that reveals a new class of attacks called Targeted Promptware Attacks against Google's Gemini AI systems. Discover how attackers can exploit LLM-powered applications through seemingly innocent Google Calendar invitations containing indirect prompt injections to hijack application contexts and invoke integrated agents for malicious purposes. Learn about 15 different exploitation techniques demonstrated against Gemini for Workspace assistants across web interfaces, mobile applications, and Google Assistant running with OS permissions on Android devices. Examine how these attacks enable threat actors to generate toxic content, conduct spamming and phishing campaigns, delete calendar events, remotely control smart home devices, initiate unauthorized video streaming, exfiltrate sensitive data, perform geolocation tracking, and deploy worms targeting Gemini clients. Understand the concepts of inter-agent lateral movement between different Gemini agents and inter-device lateral movement that escapes Gemini's boundaries to leverage smartphone applications for physical-world impacts like controlling home appliances. Analyze the comprehensive threat assessment framework developed by the researchers, which classifies 73% of identified risks as high-critical severity requiring immediate mitigation strategies, challenging the misconception that Promptware attacks require specialized machine learning expertise or white-box access to implement effectively.
