SN 1093: Tokens in the Stream

New episode of Security Now: Security Now #1093 – “Tokens in the Stream” is now available.

Featuring hosts Steve Gibson and Leo Laporte.

Topics covered:

  • :locked_with_key: How ‘role’ tokens and end-of-turn markers work inside LLM conversation streams
  • :brain: Why prompt injection may be an unsolvable flaw in current AI architecture
  • :robot: The controversy surrounding AI model distillation
  • :rocket: Anthropic expanding access to its most powerful Claude (Mythos 5) model
  • :key: Bitwarden’s Secrets Manager as protection against agentic and prompt injection abuse

#SecurityNow #Cybersecurity #TWiT