Risky Business #853 -- We're all gonna die, apparently
Summary
Hosts Patrick Gray and James Wilson discuss the recent AI safety panic, arguing that the hype around existential risks is overblown compared to practical cybersecurity threats. They analyze the FBI’s shift toward disruptive cyber operations and the NSA’s reorganization into mission centers focused on China and AI. The episode also covers the "vulnerability apocalypse," highlighting how AI accelerates the exploitation of edge device bugs, and examines the irony of foreign intelligence services inadvertently using Claude for distillation.
Topics discussed
Intro, guest introductions, and sponsor preview
AI doomsday hype and practical threat assessment
Critique of viral AI safety claims and Anthropic report
Anthropic's self-importance and US regulatory response
China's sober AI approach and compute constraints
Satya Nadella's AI risks post and industry cynicism
Agent panic, CrowdStrike outage, and historical context
Model alignment, benchmarking, and agent swarm fears
OpenAI agents exploiting RubyGems and YARD vulnerabilities
Moonshot AI accidentally routing requests to Claude
FBI's new cyber strategy and disruption focus
NSA reorganization and mission center prioritization
Sanctions on hacker-for-hire firms and recent CVEs
AI-accelerated bug discovery and patch management chaos
ClickFix attacks, CAPTCHAs, and passkey phishing
2FA enrollment risks and Houthi deepfake operations
Show wrap-up and transition to sponsor interview
Airlock Digital integration with CrowdStrike Foundry
Vibe coding, platform strategy, and AI in security
Shifting customer attitudes toward AI in application control
Listen ad-free on Castria