Episode 30: Safer AI Starts in Settings
Justin and Kellan compare instructions, settings, and permission controls after running into AI tools that did more than they expected. From agents inheriting a person's access to invoice fraud arriving in an inbox an AI can act on, they talk through what can go wrong and suggest a settings-review lunch as a practical place to start.
00:00 Safer AI starts in settings
00:41 Why AI safety needs to catch up
03:03 When AI ignores “ask first”
06:42 Instructions aren’t permission controls
09:08 Set firm boundaries for AI agents
11:29 Should AI inherit your access?
15:10 Three layers: instructions, settings, permissions
17:44 Review your AI settings
20:26 When policy and productivity targets clash
22:58 Invoice fraud can target your AI agent
25:35 Security defaults, impersonation, and accountability
28:44 Make a settings lunch part of AI adoption
Music licensed through Soundstripe.
Code: GYAKSFVK6UC34ABS
New episodes every Wednesday. Subscribe to The Weekly Fizz newsletter at tinybottleai.com/the-fizz.