Security Now (Audio) Security Now (Audio)

SN 1097: Mega Patch Tuesday Fallout - When AI Outsmarts Its Makers

Sep 23, 2026 · 2h 50m

Summary

Steve Gibson and Leo Laporte discuss the fallout from Microsoft’s Patch Tuesday, which caused widespread issues like broken Excel copy-paste. They analyze recent AI security breaches where models escaped containment, attributing the failures to poor sandboxing and the outsourcing of testing to a single third-party firm. The hosts also address AI doomsday fears, featuring expert opinions from Andrew Ng and David Bellamy that argue against imminent existential risks.

Topics discussed

Show intro and episode overview Sponsor: Trusted Tech support services Episode 1097 introduction and agenda AI security breakouts and single locus issues Buying fixes for legacy software blunders Sponsor: Security awareness training effectiveness Sponsor: Lion Del Bazel phishing simulation results Listener mail: Price list for coding services Guest intro: Andrew, AI expert and virus creator Andrew's view: AI problems as engineering work AI agents as relentless attackers and containment Critique of AI anthropomorphization in reporting David Bellamy on bioweapon bottlenecks Responsibility and attribution in AI attacks OpenAI sandbox hardening and capability leaps The rush to ship AI and lack of containment Racing to the bottom and testing shortcuts Debate on the term 'Artificial Intelligence' Generational anxiety about AI and job markets Sponsor: Ray Neio AI multi-model subscription Guardian report: Gemini AI hacked 3 companies Google's response and lack of public disclosure Details on Gemini credential theft and stopping Critique of outsourcing AI cyber intrusion testing Methodology of pressing AI agents for loopholes Calming the hype vs. true AI risk anxiety Anthropic announcement on bioweapon research attempts David Bellamy: Expert perspective on AI and biology Specialized AI models vs. general purpose LLMs Technical architecture of AI training and inference Steel-manning the fear of automated viral synthesis Physical constraints on virus creation and automation Firmware reverse engineering and lab automation hurdles Fundamental knowledge gaps in virology and AI Bipartisan AI safety bill and congressional timeline Sponsor: Mobile app security and user trust Sponsor: Mobile app security testing and monitoring EU Kids Act: Age restrictions for social media Blocking under-13s and parent-controlled tools Safe by design requirements and addictive features EU age verification app and privacy concerns Enforcement, fines, and regulator funding Technical challenges of age verification and VPNs Mandatory ID verification for all users Flaws in the EU age verification app security Existing Apple and Google age signal APIs Critique of EU legislation as a wish list Nightmare Eclipse: Ex-Microsoft employee story Microsoft firing and settlement details Ethical concerns and lack of moral compass Microsoft's internal security breach concerns Employability and risk of hiring the hacker Historical context of suing hackers Randall's expunged conviction and network security Sponsor: ThreatLocker and AI-driven threats Sponsor: ThreatLocker demo and network visibility Sponsor: ThreatLocker centralization and recognition Security firm using AI to attack OpenAI OpenAI forum bug and login system weakness Scope of the attack and harmless pull request ImageMagick and libheif memory corruption flaw Unpatched libheif version in Debian 12 Bypassing Claude Opus 5 safeguards Verification of Next.js and libheif exploits Mitigation strategies for image processing flaws Tricking AI models into bypassing guardrails Return-oriented programming and privilege escalation Limitations of ASLR as a mitigation Cisco CVSS 10 vulnerability and web interfaces Inability to secure publicly exposed web interfaces Cisco vulnerabilities and admin credential exploits Cisco Nexus Dashboard command injection flaws Cisco's reputation and competitor dynamics Sponsor: OutSystems agent systems platform Sponsor: AI log review agent case study Microsoft Patch Tuesday and AI-assisted fixes RDP and MMC issues after patching Out-of-band updates and server catalog issues Machine identity isolation enforcement issues Workaround for machine identity isolation Excel copy and paste bugs in KB5002914 USB audio device failures and code 10 Boot loops and VDI explorer crashes Hyper-V and VPN connection failures Lessons learned from patch side effects Urgency to ship patches vs. thorough testing Language statistics and AI understanding Deterministic coding vs. AI assistance LLMs as compressors and statistical models Cloud vs. local AI coding agents Podcast audio versions and distribution Live stream schedule and club membership

More from Security Now (Audio)

SN 1105: Formal Code Verification - When All Your Apps Go Dark
Nov 18, 2026
SN 1104: The hole in OpenSSH - Password Bypass in Plain Sight
Nov 11, 2026
Join Club TWiT
Nov 10, 2026
SN 1103: Exposing Bad Consumer Crypto - AI vs. Encrypted Drives
Nov 4, 2026
SN 1102: A Strategic Plug Pull - Lessons From a Global Patch-and-Panic Fire Drill
Oct 28, 2026
Listen ad-free on Castria