Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Security Now (Video)

SN 1093: Tokens in the Stream - Why LLMs are inherently insecure and prompt injection will persist

Aug 26, 20262 hr 48 minVideo
ShareXLinkedInFacebookWhatsAppEmail

Turns out, every chatbot conversation runs on a messy hack at the heart of language models, making prompt injection an unsolved - and possibly unsolvable - security threat. Steve and Leo unravel the research that explains why "roles" in AI aren't what you think they are. Understanding the controversy surrounding "AI Model Distillation" Anthropic moves to make their most powerful Mythos 5 model more widely available. Bitwarden's "Secrets Manager" offering prevents agentic and prompt injection abuse. The astounding and disturbing truth about the way conversation AI actually works Show Notes - https://www.grc.com/sn/SN-1093-Notes.pdf Hosts: Steve Gibson and Leo Laporte Download or subscribe to Security Now at https://twit.tv/shows/security-now. You can submit a question to Security Now at the GRC Feedback Page. For 16kbps versions, transcripts, and notes (including fixes), visit Steve's site: grc.com, also the home of the best disk maintenance and recovery utility ever written Spinrite 6. Join Club TWiT for Ad-Free Podcasts! Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit Sponsors: threatlocker.com/twit box.com/AI hoxhunt.com/securitynow guardsquare.com material.security

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Hosts and Guests

Hosts

  • Leo Laporte Host Feed-stated
  • Steve Gibson Host Feed-stated

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. SN 1105: Formal Code Verification - When All Your Apps Go Dark

    Nov 18, 20262 hr 42 minVideo

    What if AI could finally prove your code is bug-free? A new advance out of MIT and Google DeepMind is redefining what's possible in software verification - and it could mean a future with provably correct code, not just...

    Credit the people on this episode

  2. Join Club TWiT

    Nov 17, 20261 minVideo

    You can help support what we do at TWiT by joining Club TWiT. You'll get ad-free versions of our shows, access to the Club TWiT Discord, chapter markers, in-app transcripts, and custom programming such as Stacy's Book...

    Credit the people on this episode

  3. SN 1104: The hole in OpenSSH - Password Bypass in Plain Sight

    Nov 11, 20262 hr 38 minVideo

    From hacked French refrigerators igniting fires to a massive OpenSSH authentication bug hiding in plain sight for a decade, this week's stories expose how lapses in basic security ripple into chaos - and why AI is now...

    Credit the people on this episode

  4. SN 1103: Exposing Bad Consumer Crypto - AI vs. Encrypted Drives

    Nov 4, 20262 hr 49 minVideo

    What happens when a 27-year-old gets handed the keys to completely refactor nearly all of Cisco's code? This week's episode dives into tech's strangest management bets and the friction brewing inside one of the...

    Credit the people on this episode

  5. SN 1102: A Strategic Plug Pull - Lessons From a Global Patch-and-Panic Fire Drill

    Oct 28, 20262 hr 54 minVideo

    A newly uncovered PHP vulnerability put nearly 70% of the internet at risk, sending governments and enterprises into a frenzy as AI-powered attackers raced to exploit a flaw that had lurked unnoticed for 14 years. Hear...

    Credit the people on this episode