AI and Cybersecurity research, talk and blog in-depth articleshttps://elie.nethttps://elie.net/favicon/favicon.svghttps://elie.net/favicon/favicon.svg0099ffhttps://elie.net/publication/facade-high-precision-insider-threat-detection-using-deep-contextual-anomaly-detectionhttps://elie.net/publication/facade-high-precision-insider-threat-detection-using-deep-contextual-anomaly-detectionFacade detects insider threats by learning the context of everyday activity, achieving the precision needed to protect a large corporate environment.Wed, 12 Aug 2026 12:00:00 GMThttps://elie.net/publication/exploitgym-can-ai-agents-turn-security-vulnerabilities-into-real-attackshttps://elie.net/publication/exploitgym-can-ai-agents-turn-security-vulnerabilities-into-real-attacksExploitGym measures whether AI agents can turn real vulnerabilities into working exploits across applications, V8 and the Linux kernel.Mon, 11 May 2026 12:00:00 GMThttps://elie.net/publication/droidcct-cryptographic-compliance-test-via-trillion-scale-measurementhttps://elie.net/publication/droidcct-cryptographic-compliance-test-via-trillion-scale-measurementDroidCCT tests Android cryptography at ecosystem scale, analyzing trillions of samples from half a billion devices to identify implementation weaknesses.Mon, 08 Dec 2025 12:00:00 GMThttps://elie.net/publication/evaluating-the-robustness-of-a-production-malware-detection-system-to-transferable-adversarial-attackshttps://elie.net/publication/evaluating-the-robustness-of-a-production-malware-detection-system-to-transferable-adversarial-attacksA study of Gmail malware detection shows how adversarial attacks on one ML component affect the entire pipeline, and evaluates a deployed defense.Wed, 19 Nov 2025 12:00:00 GMThttps://elie.net/publication/integrating-large-language-models-into-security-incident-responsehttps://elie.net/publication/integrating-large-language-models-into-security-incident-responseA study with security analysts examines how large language models can help summarize incidents, where they make mistakes and when collaboration helps.Mon, 11 Aug 2025 12:00:00 GMThttps://elie.net/talk/facade-high-precision-insider-threat-detection-using-contrastive-learninghttps://elie.net/talk/facade-high-precision-insider-threat-detection-using-contrastive-learningThis talk presents FACADE, a novel self-supervised AI system used by Google to detect insider threats with high precision. FACADE uses a contrastive learning strategy trained solely on benign data, achieving a false positive rate below 0.01%.Thu, 07 Aug 2025 09:00:00 GMThttps://elie.net/talk/autonomous-timeline-analysis-and-threat-hunting-an-ai-agent-for-timesketchhttps://elie.net/talk/autonomous-timeline-analysis-and-threat-hunting-an-ai-agent-for-timesketchAutonomous timeline analysis and threat hunting with the Sec-Gemini digital forensic agent, and the Timesketch AI Panel designed to make its findings actionable.Wed, 06 Aug 2025 09:00:00 GMThttps://elie.net/publication/supporting-the-digital-safety-of-at-risk-users-lessons-learned-from-9-years-of-research-and-traininghttps://elie.net/publication/supporting-the-digital-safety-of-at-risk-users-lessons-learned-from-9-years-of-research-and-trainingLessons from more than nine years of research and training at Google inform how to support the digital safety of people facing elevated risks.Sat, 14 Jun 2025 12:00:00 GMThttps://elie.net/publication/supporting-human-raters-with-the-detection-of-harmful-content-using-large-language-modelshttps://elie.net/publication/supporting-human-raters-with-the-detection-of-harmful-content-using-large-language-modelsFive design patterns show how large language models can assist human content reviewers, improving capacity and the detection of harmful material.Mon, 12 May 2025 12:00:00 GMThttps://elie.net/publication/magika-ai-powered-content-type-detectionhttps://elie.net/publication/magika-ai-powered-content-type-detectionMagika uses a compact neural model to identify file content types accurately and quickly, supporting attachment scanning and malware analysis.Sat, 26 Apr 2025 12:00:00 GMT