AI hates CAPTCHAs - PSW #944
Summary
Hosted by Larry Pesci with guests Sam Bown and Lee Neely, this episode explores the limitations of AI, noting that models struggle with captchas and can be easily "jailbroken" via open-source weights. The discussion covers the privacy concerns of Apple Watch's new always-on transcription feature, comparing it to Microsoft Recall, and highlights how AI agents can be used for security research, such as reverse-engineering Android vulnerabilities. The hosts also debate the hype surrounding AI risks, arguing that current models are merely advanced search tools rather than conscious entities, wh…
Topics discussed
Intro: AI, Captchas, and Guest Lee Neely
Sponsor: Identity-based attack detection
AI Model 'Obliteration' and Chinese Distillation
AI Agents Struggling with Captchas
Sponsor: Zero Trust and ThreatLocker
Data Center Infrastructure and Solar Mitigation
China's AI Strategy and Captcha Obstacles
AI Scrapers, Rogue Agents, and Human Oversight
Pixel Modem CVE and AI Exploit Generation
AI in Financial Fraud and Pattern Recognition
Rogue AI Agents and Cloud Resource Limits
Book Signing and ESP32 Hardware Projects
Apple Watch AI, Secure Enclave, and Privacy
Eavesdropping via Wi-Fi Signals and Walls
Hidden AI Prompts in Legal Filings
Quantum Computing, Qubits, and Post-Quantum Crypto
Flock Camera Hacks and Police Database Sloppiness
VBScript Malware and Threat Hunting
Cloud vs Local Infrastructure and Recycling
Passkeys Adoption and User Resistance
CRA Compliance and Vulnerability Reporting
Cyber Attack Costs and Insurance
Microsoft Employee Lawsuit
Check Point and GitLab Vulnerabilities
Background Checks and GitLab Patching
Strait of Gibraltar Ship Hacks
Parallels Desktop Flaw and ARM Windows VMs
Listen ad-free on Castria