SN 1094: AI Patching Shortcomings

New episode of Security Now: Episode #1094, “AI Patching Shortcomings,” is now available.

Hosts: Steve Gibson and Leo Laporte

Topics covered:

  • :robot: New research showing frontier AI models fail at vulnerability remediation, with nearly half of AI-generated code containing critical flaws
  • :shield: A possible approach to preventing prompt injection abuse and whether better harnesses could help
  • :satellite_antenna: Clear evidence of malicious intent in Chinese-made routers
  • :open_mailbox_with_raised_flag: Listener feedback on AI terminology, including whether AI truly “thinks,” “rotating credentials,” and a proposal to add unpredictable hashes to role tags
  • :floppy_disk: A before-and-after SpinRite graph showing improved SSD performance, plus what Claude thought of last week’s podcast

#SecurityNow #Cybersecurity #TWiT