Defensible Application Security for the Artificial Intelligence Era
summary
Claims that AI breaks traditional trust: facial recognition tests show high mismatch rates—32% mismatch (Amazon) and 24% mismatch (Google Cloud) in a 2017 dataset and ACLU finds 28 U.S. Congress photos falsely matched to a criminal database (~5%). Resume‑screening AI proves gender bias. A proprietary algorithm extracts heart‑rate variability from video and closely tracks oximeter readings. Zero‑day market prices hit $1.5M in 2017, suggesting a lucrative market for human‑vulnerability signals and the need to test models against adversarial samples.
AI-written summary · about
chapters (7)
- 0:00Start
- 7:50What About This?
- 9:23Trust From A Neuroscience Point Of View
- 13:37Facial Recognition Is Getting Better
- 21:39Heart Rate Variability (HRV)
- 28:31Price Of Zero Days For Mobile (2017)
- 29:04What Should Be The Price Of Human Vulnerabilities?
related talks
26mAI is not a god that we can trust
17mAI Safety in the Generative AI Era
57mAI Applications Security Puzzle
42mIs your AI plotting to kill you? Why AI and Machine Learning is dangerous
26mSafeguarding AI: How to Manage AI and its Potential Risks
40mEthical AI: Addressing Bias & Algorithmic Fairness
50mSecurity as Part of Responsible AI: At Home or At Odds?
01h 05mNavigating the Security and Privacy Landscape of Modern AI