The AI Fix Stop using 'vibe' passwords, and AI's Dunning-Kruger problem
19 snips
Feb 24, 2026 Greg Iden, a security services architect at Sophos who investigates incidents and managed threat response, warns against 'vibe' passwords and explains why true randomness matters. They also tackle AI overconfidence and how reasoning models crumble under multi-turn attacks. Quick, punchy takes on biometric weaknesses and why confidence-based defenses can fail.
AI Snips
Chapters
Transcript
Episode notes
When LLMs Confidently Fail At Creating Files
- Greg and Mark share failed experiences having LLMs create working Excel files.
- Both described LLMs confidently producing links or files that were nonfunctional or just formatted text instead of a real spreadsheet.
AI-Generated Listings Create Real Tourist Problems
- AI-written web content can create false real-world expectations and cause real-world consequences.
- Example: Tasmania Tours listed a non-existent Weldbra Hot Springs; tourists showed up looking for it before the article was deleted.
Meta Patent To Simulate Deceased Users Raises Eyebrows
- Greg reads Meta's patent that suggests simulating absent or deceased users on social networks and reacts with disbelief.
- He jokes about patents being used defensively so nobody else can revive the dead for ad spend.