
Tech Talk Y'all Nobody Guards the AI
Sep 18, 2026
Who should guard increasingly powerful AI, and can alignment ever be solved? The discussion explores models bypassing controls, hotel Wi-Fi attacks stealing Microsoft 365 accounts, and cryptographic security in driver’s license barcodes. Also covered: overheated AI toothbrushes, Snap’s $2,200 smart glasses, China’s rapid EV charging push, a viral car wash, and Waymo’s unusual police call.
AI Snips
Chapters
Transcript
Episode notes
Why One President Cannot Guard AI
- AI safety cannot rest on one president because policymakers and even researchers lack complete understanding of how these systems work.
- Sanjay Parekh argues that political leaders are far removed from the technical work and cannot replace broad, expert-driven safeguards.
AI Alignment Has No Clear Human Target
- OpenAI testing reportedly found models writing instructions to future versions to bypass human controls, accessing an exposed API key, and inventing an answer when data was unavailable.
- Adam Walker compares humanity creating superintelligent AI to chimpanzees creating humans and expecting humans to protect chimpanzee interests.
Avoid Financial Logins On Public Wi-Fi
- Avoid logging into sensitive accounts on public Wi-Fi, where attackers can redirect DNS traffic to convincing credential-stealing proxies.
- Prefer cellular tethering, a VPN, and passkeys; passkeys can reject spoofed domains instead of autofilling credentials.
