CAIS 2026 | Securing Gemini from Cyberattack Misuse

An inside look at how Google DeepMind detects and defends against attempts by threat actors, including state sponsored groups, to misuse Gemini across the stages of a cyberattack, from reconnaissance and phishing to malware development and vulnerability research, and the safeguards built in response. Speakers & affiliations: Raluca Ada Popa, Gemini security lead at Google DeepMind and Professor of Computer Science at UC Berkeley Session: Operational Experience 3 · Friday, May 29 Key terms: AI misuse, threat intelligence, adversarial AI, cyberattack lifecycle, Gemini security, safety classifiers, CodeMender (vulnerability detection and repair) ACM Digital Library: N/A. Invited operational experience talk, one of three such talks, distinct from the peer reviewed proceedings papers, no individual DOI. (Proceedings: 10.1145/3786335.)