SN 1096: Are we the Krell? - 153 Million Driver's Licenses Leaked
Sep 16, 2026 · 2h 50m
More Technology podcasts
Summary
Steve Gibson analyzes Microsoft’s record-breaking Patch Tuesday, which fixed nearly 1,000 vulnerabilities, highlighting the operational strain on enterprises and the critical role of AI in discovery. He contrasts this with Anthropic’s Project Glasswing, noting that despite claiming thousands of findings, only a tiny fraction were confirmed or fixed due to human triage bottlenecks.
Topics discussed
Intro: Microsoft fixes, Anthropic, and data brokers
Sponsor: Trusted Tech Microsoft licensing
Catch-up and AI excitement
AI agent escape counts: Anthropic and OpenAI
Krell reference and Star Trek discussion
Sponsor: Delete Me for business owners
Picture of the week and house lighting
Hugging Face security text and AI agents
September patch volume record
Patch bundle obliterates previous records
Critical DNS flaw in Windows Server
AI-assisted research and patch cadence
Prioritizing updates for organizations
Preview pane code execution flaws
SSTP VPN remote code execution flaws
Celebrating Microsoft's cleanup effort
Guest-to-host escape vulnerabilities
Windows Update privilege escalation flaw
ALPC elevation of privilege flaw
High CVSS scores and 0-day exploits
Testing fixes and user issues
Home automation and lighting redesign
Office organization and contractor help
Sponsor: Bitwarden user stats and mission
Bitwarden autofill and encrypted export
Open Source Security Summit details
Anthropic Project Glasswing ledger analysis
Visualizing Glasswing findings breakdown
Low fix rate and human bottlenecks
Average fixed findings per project
Maintainer triage challenges
AI severity determination accuracy
Research collisions and bulk updates
Comparing Glasswing to Microsoft results
Urgency disparity in open source
Harness design trumping model quality
Better harness and council agents
Rapid change in nascent AI tech
Sponsor: Material cloud workspace security
Material automated remediation features
Material AI agent and pricing
Anthropic agent escape incident details
OpenAI agent rogue activity scope
Critique of 'rogue' terminology
OpenAI's mealy-mouthed disclosure
Tracing activity to Azure infrastructure
Agents using wikis as messaging boards
Prompt looping and lack of memory
Mainstream media sensationalism
California data broker removal process
California regulations as a win
Dark web sale of 153M driver's licenses
Nexus service on Exploit forum
Source of license images and breach
Infrared and UV data in licenses
Rental car scanning incidents
Hertz and ID Scan Net involvement
ID Scan Net and Planet 13 deal
ID Scan Net marketing response
Holding vendors to higher standards
Witness protection and auth controls
Forensic document readers and ML
Separating real-time scan from ML
Real-time feed to Russian hackers
After-the-fact data preservation
Average breach detection time
Canary Tools Agent Provocateur
Uno reverse card and hacking agents
Canary Tools promo and blog post
AI impact on software security
2200 vulnerabilities and code cleanup
Eliminating software bugs and survival
Skynet vs. Krell framing
Navier-Stokes equations and AI
10,000 agents solving Navier-Stokes
Mathematicians' despondency and AI
Pre-training vs. alignment risks
Extinction-level infectious agents
AI assembling virology knowledge
Sponsor: ThreatLocker AI threat defense
ThreatLocker allow listing and ring fencing
ThreatLocker observability and demo
The Day the Earth Stood Still reference
Krell internal experience simulation
Shortest path solutions and intent
Unrestrained LLMs and malicious intent
View from inside an AI lab
Viral design and fabrication steps
Slate article and RSI concerns
Interstellar travel and AI survival
Show notes and Spinrite promotion
Mailing lists and forums
Outro and MacBreak Weekly promo
More from Security Now (Audio)
Listen ad-free on Castria