Securing AI on the Endpoint, Anthropic Dreams of AI Replicants, and the News - Michael Jenkins - ESW #481 — Security Weekly Podcast Network (Audio)

Host Adrian Sanabria interviews ThreatLocker CTO Michael Jenkins on securing AI agents on the endpoint using zero trust and ring-fencing strategies. The hosts then debate Anthropic’s recent research on self-replicating AI swarms, questioning the feasibility of running inference on IoT devices like smart TVs. The episode concludes with a news segment covering enterprise security topics, including Microsoft’s massive patch backlog, crypto wallet supply chain issues, and the industry’s most costly habits.

Topics discussed

Introduction and episode overview Sponsor: Network Security Virtual Summit Casual chat on desktop clutter The ubiquity of AI in daily workflows AI capabilities: computer vision and web automation AI accessing sensitive data: banking and healthcare Personal AI assistants and data privacy concerns Promoting content and podcast boundaries Hybrid AI models and local execution Sponsor: ThreatLocker and password management Vibe coding a Pebble watch app The annoyance of frequent permission prompts Managing AI permissions and access control Applying zero trust principles to AI Zero trust: saying yes in specific circumstances Employee understanding of AI restrictions Development environment for the watch app Ring fencing concept in zero trust security Examples of ring fencing: OpenAI and SolarWinds Testing ThreatLocker on a production machine Reducing permission prompts with ThreatLocker Package-based approval vs individual executables Teasing an upcoming ThreatLocker announcement Sponsor: ExaForce and transition to topic segment Topic: AI replicating itself Critique of AI replication research assumptions Welcome to co-host Hyman and discussion threads Academic formatting and LaTeX in AI research Lack of peer review and evidence in AI papers Target audience and practical concerns of AI replication Impact on consumer devices like smart TVs Counter-engineering and timing of AI replication Theoretical feasibility and comp sci perspective Technical merit of resource-constrained AI models Distinction between ML models and LLMs Low resource vs high resource environments CISO credibility and alarmist claims Academic paper vs blog post publication TLDR for the board: no action required Sponsors: Bold Security and Travelers Cyber Threat Report Vibe Check segment introduction Focus on external threats vs internal misconfigurations MFA rollout vs AI automated red teaming CISOs buying innovations too early Is security reinventing itself or stuck in 2010? Management challenges in large-scale security Conclusion of Vibe Check segment Funding news: Rain Security and others Certifying human-made content in the AI era Pride in writing by scratch vs AI assistance AI as performance-enhancing drugs in creative work Challenges of AI watermarking at scale Value collapse of the written word Story: AI joining a group chat automatically Meta's automated AI-based account issues Recovering a locked account via internal contacts Wiped Google and Apple developer accounts Longevity of cybercriminal groups Hostile takeovers in the carding scene UK Minister for AI speech analysis Survivorship bias in breach data Abstracted political predictions vs technical reality Mobile security evolution and Pwn2Own AI security path: mobile vs internet Browser security history and Pwn2Own Significance of AI security attention Hardware hacking: Ledger wallets with LTE modems Crypto recovery industry and data recovery Identifying compromised Ledger hardware Specific model identification of compromised devices Geographic limitations of the Ledger issue Assessing risk for existing Ledger users Miscellaneous tech news and food storage Using AI bots to find and report default passwords Risks of attempting to do the right thing Ransomware breach at Del Haze grocery stores PR challenges during a grocery store outage Predicting patch Tuesday volume trends Creating frenzy around non-critical issues Testing websites in sandboxes before cloning Closing remarks and next week's preview

Episode summaries and topics are generated from the audio by Castria — a free podcast app that skips ads automatically. IP Geolocation by DB-IP

Securing AI on the Endpoint, Anthropic Dreams of AI Replicants, and the News - Michael Jenkins - ESW #481

Summary

Topics discussed