Fighting AI with AI
summary
Shows that deepfakes are already practical—one audio clone takes 72 hours to create on powerful GPUs and fools 90% of people, so defenses must also use AI. Uses Behavioral Signals’ voice detection idea: build a behavioral profile from 75–100 attributes, synthesize short fake utterances to harden models, and detect mismatches with as little as 10–20 seconds of audio. Claims high‑90s detection accuracy, language‑agnostic scalability to terabytes, and needs real‑time, provenance and multimodal work.
AI-written summary · about
chapters (15)
- 0:00AI’s duality and deep fakes
- 1:27Deep fakes threat growth
- 3:35Deep fakes with GANs
- 5:00Defining reality challenge
- 10:45Detecting deep fakes now
- 13:54Traditional detection failure
- 15:15Behavioral profiling for deep fakes
- 23:46Deep fake detection future
- 27:37AI arms race
- 31:10Regulation and responsibility
- 35:23Future of deep fakes
- 36:03Behavioral profiling accuracy
- 40:18Updating behavioral profiles
- 42:33Ethics and military use
- 46:04Global AI regulation
related talks
30mAI-Generated Synthetic Media, aka Deepfakes and its impact
27mTruth or Fiction The Art of Unveiling DeepFakes in Text, Audio, Image
30mReality Under Attack: Just How Serious Is the Deep Fake Threat?
26mDeepfakes, audiofakes, and the future of media
36mFabricated Realities: AI Personas and Simulacra and the Threats They Pose
39mSynthetic Trust: Exploiting Biases at Scale
36mDemocratized Deception: How Threat Actors Are Weaponizing AI to Scale Cybercrime
45mFUD vs. Reality: Bracing for an AI-Powered Threat Landscape