How Large Language Models Are Reshaping the Cybersecurity Landscape
summary
Argues that large models (Gemini, GPT/Claude equivalents) both lower attacker barriers and scale defenders by deploying a LLM rater for content moderation, using retrieval-augmented generation to lift accuracy (RAG pushes scores into the 90s, election misinformation ~99%), and combining multimodal checks with Gmail telemetry (70% image+text malicious docs). Reports engineering experiments: fuzzers guided by LLMs find new bugs, incident summaries are 51% faster, and automated patching achieved ~15% success on 1,000 real bugs.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:53Large models today
- 4:42Offensive AI risks
- 7:13Deepfake threats
- 10:35Black market AI use
- 13:26Threat landscape
- 15:32Defensive AI shift
- 16:35AI for content review
- 23:23Retrieval augmented generation explained
- 29:33Malicious document analysis
- 34:11AI for code security
- 41:43AI in incident response
- 43:51Future outlook
- 45:15Q&A
related talks
13mIntegrating Large Language Models into Security Incident Response
23mSecurity and auditing tools in Large Language Models (LLM)
31mHacking with Words...
47mCurrent Security Threats in Large Language Model Applications
39mThreat Modeling Using LLMs
01h 04mCan you trust your (large language) model?
44mI Abuse Your LLM Live with 10 Security Vulnerabilities
47mHarnessing LLMs for Next-Level Security Analysis