Oct 3, 2026 · 4h 5m
AI agents expose the gap between autonomy and accountability
AI Ruined The Deal And Gave Away His Address - WAN Show October 2, 2026
The episode tests whether AI can safely act in the real world while examining the fragile systems, incentives, and hardware surrounding it.
- 1AI agents remain risky when platforms limit access, users retain responsibility, and basic safeguards fail.
- 2Local AI promises control and customization, but hardware costs, permissions, and reliability complicate the case for running models at home.
- 3Across games, smart homes, gambling, and cybersecurity, convenience repeatedly collides with durability, transparency, and responsible design.
Don't miss
The hosts dissect an AI Marketplace agent that accepted a low offer and exposed a seller’s pickup address, making automation’s accountability problem concrete.
The brief
Meta’s AI agent reportedly accepted a low Facebook Marketplace offer and disclosed a seller’s pickup address, raising a basic question: who is accountable when automation acts badly?
Linus Sebastian’s local assistant, Sventlana, shows why useful agents need permissions, integrations, and judgment—not merely a capable model behind a chat window.
The hosts weigh local models, smart-home automation, and AI-assisted cybersecurity against hardware costs, brittle connections, expanding attack surfaces, and the difficulty of reliable control.
Beyond AI, the episode links Destiny 2’s restored content, Sony’s PS5 Pro lottery rules, GPU restrictions, and game-currency regulation through a shared concern with access and permanence.
The broader takeaway is skeptical but practical: technology becomes valuable when it reduces friction without quietly transferring risk to users, communities, or future maintainers.
Featuring
Linus Sebastian
PlayStation 5
Apple Inc.
Nvidia Corporation
World of WarcraftListen to the full episode and explore every guest, topic, and moment on PodLume.