Pierce Freeman and Richard Diehl Martinez examine who should control increasingly capable AI systems, starting with Apple's June announcement that Siri AI would be delayed on iPhone and iPad in the EU. They favor Apple's security concerns and use the dispute to discuss the risks of assistants that can access personal information and act across apps. Apple's interpretation of the Digital Markets Act is contested, rather than an established requirement for unrestricted access without consent.
Their explanation of an agent harnessAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction. separates the language model making decisions from the tools and permissions surrounding it. They argue that introducing more integrations can expand the attack surface and that privacy depends on the whole systemData privacy governs how personal, confidential, or sensitive information is collected, used, shared, retained, and protected in AI systems., not just the model. The conversation is a conceptual risk discussion, not proof that all third-party tools are unsafe or a closed ecosystem cannot leak data.
The hosts then discuss Fable 5's capabilities and speculate about how safeguards might restrict sensitive tasks or assistance to competing AI developers. Their most concrete example is an automated research loop: propose a change, run it, evaluate it and iterate. They distinguish fast, measurable kernel optimization from broader model research whose experiments and goals are harder to evaluate.
Recorded during the June suspension of Fable 5 and Mythos 5 access, the episode questions the government's growing roleGovernance defines decision-making authority, rules and accountability within a system or organization. and whether publicity around restrictions also benefits a model provider. The hosts do not establish a hidden commercial motive or the exact reason for every restriction. Subsequent restoration means this portion should be read as a snapshot of the June dispute, not a continuing claim that Fable 5 is unavailable.
The final segment considers autonomous weapons, drawing a distinction between automating execution of a human decision and allowing a system to select a target itself. The hosts focus on accountability, meaningful human controlHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work. and international coordination. Their more extreme persistent-drone scenarios are speculative, and the discussion does not demonstrate that such systems already exist or settle the law governing them.
Watch on YouTube




